Image: Zambia’s 2027 Revenue Target, the ZMW30.73bn Above the 2026 Pace and What Fills It

Canary Compass will move to paid subscriptions soon. Founding members put a question to Dean every month and join a quarterly call. Founding membership is limited to 50 people and closes at launch. Pledge now; you pay nothing until paid publishing begins. Friday Reflections will stay free. See every plan and what it includes. Institutional licences: dean@canarycompass.com

Chart data: founding members, institutional seats and Desk licence readers receive the values behind this Note’s charts on request.

Zambia’s 2027 Budget Address, delivered on 6 November, plans a deficit of 3.0 per cent of GDP. Minister of Finance and National Planning Dr Situmbeko Musokotwane now projects the 2026 deficit at 5.9 per cent, against 2.1 planned. The Address sets spending, in Zambian kwacha (ZMW), at ZMW284.8bn including external principal repayments, domestic revenue at ZMW233.7bn and grants at ZMW9.2bn. It finances the rest with ZMW28.9bn of net domestic borrowing and ZMW13.0bn of foreign loans, and assumes real growth of 4.9 per cent.

This Note tests the Budget line by line against what the Treasury collected in the first eight months of 2026, and answers four questions. How much of the 2026 shortfall came from the exchange rate, which the Address does not name? What exchange rate does the 2027 Budget stand on, when it prints none? Which revenue lines hold at the current exchange rate with the fuel taxes back, and which do not? Who buys the government’s paper if revenue falls short?

PAYWALL LINE. From the paid launch, free readers stop here. The summary box, our numbers and the verdict below are for subscribers. Until then, this Note is free in full.

1. What 2026 Left Behind

Musokotwane gives 2026 two causes (Address, p.2). The first is the suspension of value added tax (VAT) and excise duty on fuel, which he says “largely” explains the shortfall. The second is food security spending. Inflation fell to 6.3 per cent in October, which he credits to the maize harvest and the stronger kwacha (p.2). Parliament approved a supplementary appropriation of ZMW26.3bn on 11 May (Supplementary Appropriation Act No. 78 of 2026, assented on 4 June). A second, of ZMW19.9bn, is at third reading (Supplementary Appropriation (2026) (No. 2) Bill, N.A.B. 79 of 2026). The Bill gives ZMW14.7bn to National Food Reserves Management (Head 89). Musokotwane told the House on 22 October that the money pays farmers for maize sold to the Food Reserve Agency. He said ZMW18.0bn of the supplementary would come from additional domestic borrowing (as reported, Times of Zambia, 23 October 2026). The 2027 Estimates of Revenue and Expenditure record 2,743,393 tonnes of maize bought in 2026 to 30 June against a target of 1,000,000 (Head 89, p.820). International Monetary Fund (IMF) staff agreed a new 36-month Extended Credit Facility (ECF) of about USD1.5bn on 9 October (IMF Press Release 26/331).

The revenue record supports his first cause in part and shows a larger one he does not name, the exchange rate. From January to August, revenue and grants reached ZMW127.4bn against the Ministry’s own profile of ZMW146.9bn: a 13.3 per cent shortfall (the Green Paper, the Ministry’s September 2026 draft of this Budget, Table 3, p.27). The Monthly Economic Indicators (MEI) for August 2026 carry the same outturn. Revenue alone fell ZMW14.8bn short; delayed grants were the other ZMW4.7bn.

The Government suspended excise and VAT on fuel from April. From fuel volumes and the Energy Regulation Board’s monthly price build-ups, we estimate the cost at ZMW6.3bn to ZMW7.1bn from April to August: ZMW6.7bn at the midpoint, a little under half of the ZMW14.8bn. Import VAT, the largest budgeted line at ZMW42.5bn, ran ZMW11.5bn below eight-twelfths of its budget. The fuel relief explains about ZMW5.5bn of that and the exchange rate about ZMW6.3bn, at the 2026 Budget’s rate of 26.00 (section 2). Together the two estimates cover the whole gap.

On the lines the Budget collects in dollars, we estimate the kwacha’s strength cost about ZMW20.6bn, more than the whole shortfall. Mineral royalty and mining company tax lost almost half of that, and higher copper prices more than replaced it: the two lines ran ZMW5.3bn above eight-twelfths of their budgets. The other ZMW10.8bn, on import VAT, customs duty, export duties and grants, had no such offset. On our estimates, fuel (ZMW6.7bn) and the exchange rate outside mining (ZMW10.8bn) together cost ZMW17.5bn. Against eight-twelfths of budget, revenue and grants ran ZMW18.3bn short in all and ZMW23.6bn short outside mining. The two estimates account for about three-quarters of the second figure, and domestic VAT, ZMW6.1bn short, is the largest of the other shortfalls (section 4). The Address credits the stronger kwacha with bringing inflation down, but leaves it out of the revenue account. Its own draft names “exchange rate appreciation” among the causes of the miss on import VAT and road tolls (Green Paper, paras 101 to 103, pp.24 to 25).

The public record supplies only part of the arithmetic behind Musokotwane’s deficit projection. By August, net domestic financing had reached ZMW39.0bn (MEI, August 2026, Table 1.0) against a profile of ZMW15.4bn (Green Paper, Table 3, which prints a preliminary ZMW38.6bn). The Bank of Zambia (BoZ) lent ZMW9.6bn of it towards the June Eurobond buyback. The African Development Bank lent a further ZMW10.4bn for the same buyback, which the Ministry books as foreign financing (MEI, August 2026, Table 1.0; Green Paper, para 114). Even with that loan, foreign repayments, which include the buyback (Green Paper, paras 109 and 114), exceeded foreign loans by ZMW13.8bn, so net financing reached ZMW25.2bn. The deficit to August was ZMW27.2bn; a change in balances covered the other ZMW2.0bn (MEI, August 2026, Table 1.0). At 5.9 per cent of the ZMW923.7bn the 2026 Budget assumed, the full-year deficit is near ZMW54.5bn, against the ZMW19.4bn the approved Budget financed. The last four months therefore carry about ZMW27.3bn of it. The Supplementary No. 2 Estimates and the amended 2026 borrowing plan, which the House adopted on 3 November (as reported, News Diggers, 4 November 2026), will show the split.

2. The Kwacha Across the 2026 Budget

The 2026 Budget publishes no exchange rate: the Ministry’s assumptions table, as the Zambia Revenue Authority (ZRA) presented it on 29 September 2025, carries none (slide 7). It sets copper at USD8,897 a tonne and output at 1,069,055 tonnes. The 2026 Annual Borrowing Plan, issued with the Budget, supplies the rate. It plans loan disbursements of USD494.98m as ZMW12.87bn and external principal of USD573.04m as ZMW14.90bn: USD/ZMW 26.00 on both. The approved Budget carries two of the Plan’s loans at the same rate (MEI, August 2026, Table 1.0; Plan annex). This Note uses 26.00 as the 2026 Budget’s rate. The Ministry’s Debt Statistical Bulletin schedules more dollar principal for the same year: USD747.66m in the Quarter 2 2025 edition and USD678.16m in the End Quarter Four 2025 edition (Table 15). On those dollars the Budget’s ZMW14,898m of external principal would read 19.93 and 21.97.

The kwacha’s January-to-August average of 18.91 (MEI, August 2026, Table 4.0) ran 27.3 per cent stronger than the Budget’s rate, which gives section 1’s ZMW20.6bn with each line’s dollars held fixed.

A line counts as dollar-linked only where the Budget collects or pays a dollar amount that converts to kwacha. In revenue those are import VAT, customs duty, export duties, mineral royalty, mining company tax and grants; in spending, external interest, foreign-financed spending and overseas allowances; in financing, external principal and loan proceeds. Lines with some dollar content and no published split, non-mining company tax and withholding tax among them, stay out, so the shares below are floors. Copper explains the offset on the mining lines (section 1): it averaged USD13,322 a tonne from January to August (BoZ fortnightly statistics), 49.7 per cent above the assumed price. The Address projects 2026 output at 930,000 tonnes (para 24), 13.0 per cent short of the assumed output, so price beat the plan by far more than volume missed it.

On that classification, dollar amounts make up 46.1 per cent of revenue and grants in the 2026 Budget and 10.4 per cent of spending. With excise, a wider measure, the revenue share reaches 52.6 per cent. A kwacha 10 per cent stronger therefore leaves the domestic market ZMW7.4bn more to carry. Table 1 sets that arithmetic beside the January-to-August estimate: ZMW16.7bn of extra domestic financing, 1.8 per cent of the ZMW923.7bn the 2026 Budget assumed.

3. Where the Budget Prices the Kwacha

Neither the 2027 Address nor the Estimates published with it prints an exchange rate. The 2027 Annual Borrowing Plan again supplies it. The Plan puts 2027 foreign loans at USD605.99m, and the Address books them at ZMW13,029m (p.20). The Plan repays USD550.52m of external principal, and the Green Paper provides ZMW11,836m for it (fiscal framework, p.40). Both pairs give USD/ZMW 21.50, the rate this Note uses for 2027. That principal sits inside the ZMW16,764m of external debt service the Address and the Estimates carry (p.17; Head 99). The Debt Statistical Bulletin again schedules more: its Quarter 2 2026 edition (updated) projects 2027 principal of USD658.03m (Table 15), on which the same kwacha would read 17.99.

Every monthly average from January to August was stronger than 21.50; the weakest, January’s, was 20.09 (MEI, August 2026, Table 4.0). The IMF’s October figures imply USD/ZMW 20.48 for 2027, nominal GDP of ZMW933.5bn over USD45.58bn (World Economic Outlook (WEO), October 2026), 4.7 per cent stronger than the Budget’s rate. The market traded at 20.38 at the BoZ 12:30 fixing on 10 November (BoZ interbank rates; Chart 1), 5.2 per cent stronger.

The Address keeps a flexible exchange rate and will keep building reserves, lifting cover above five months of imports over the medium term, partly by buying gold at home (pp.13 to 14). Cover fell to 4.4 months at end-June from 5.2 at end-March (Debt Statistical Bulletin, Quarter 2 2026 edition, section 2.2); reserves stood at USD6.1bn at end-August (Address, para 15). Where the BoZ builds them by buying dollars at home, it takes supply out of the market, which works against a stronger kwacha.

In 2026 the current account ran a first-half surplus of USD416m, against a USD840m deficit a year earlier, as copper exports rose 54.0 per cent to USD6.1bn. Non-residents added USD604m of Zambian debt securities in the first quarter (BoZ Balance of Payments Statistics, Q2 2026). A kwacha at the Budget’s 21.50 would take less support from those two flows, more demand for dollars or more buying by the BoZ. The limit planned for non-residents at bond auctions from 2027 narrows the second flow (section 6).

The two borrowing lines, ZMW41.9bn together, are not the deficit: they also repay external principal falling due, which the Green Paper splits out. Spending excluding that principal, less revenue and grants, leaves a deficit of ZMW30.04bn, 3.0 per cent of the Green Paper’s GDP (section 5). Foreign loans exceed the principal by about ZMW1.2bn, so net domestic borrowing of ZMW28.85bn funds 96.0 per cent of the deficit.

The market sits 5.2 per cent stronger than the Budget’s rate, where the January-to-August 2026 average sat 27.3 per cent stronger than that year’s. Hold the kwacha at the 10 November fixing for all of 2027 and the Budget’s dollar lines collect about ZMW5.6bn less. The deficit runs about ZMW4.3bn wider, 0.4 per cent of GDP.

In 2027 slightly less of the Budget is dollar-linked than in 2026 (section 2), and more of that on copper. Four lines the Address prints are fully dollar-linked: customs duty, export duties, mineral royalty and grants. They make up 19.4 per cent of revenue and grants. Adding import VAT and mining company tax, which the Estimates print separately, lifts the figure to 44.6 per cent, or 50.2 per cent with excise. Within the dollar-linked total, the copper-linked lines, mineral royalty plus mining company tax, rise from 31.1 per cent in the 2026 Budget to 41.8 per cent in 2027. Table 2 shows what a 10 per cent stronger kwacha would cost; the 10 November fixing is already half that move.

4. What 2027 Asks of Revenue: The Repricing Test

The method takes each line’s January-to-August 2026 collection and scales it to a full year (Table 3). Unadjusted, that pace is ZMW186.2bn, and the ZMW233.7bn target sits 25.5 per cent above it. Two adjustments make the pace fair to 2027. First, fuel: excise came back in October and VAT comes back on 1 December (Energy Regulation Board price review, 30 September 2026), so 2027 collects both all year. The eight months hold only three months of fuel tax, January to March. Scaled to a year that is 4.5 months, so the method adds the missing 7.5 at the April-to-August cost of ZMW1,343m a month: ZMW10.1bn. Second, the kwacha: repricing the dollar-linked lines from the January-to-August average to the 10 November fixing lifts them 7.8 per cent. Adjusted, the pace is ZMW202.93bn against the target’s ZMW233.66bn. The targets are the lines of the Estimates (pp.xv to xl), which split VAT and company tax where the Address does not. The scaling is straight-line, so the figures are our estimates. The Ministry’s own profile to August expects 66.7 per cent of the year’s tax revenue and 67.3 per cent of domestic revenue (Green Paper, Table 3), so seasonality moves the total pace little.

Lines carrying a third of the target clear the test (Chart 2): each needs less than the 8.6 per cent by which the Budget grows the economy in kwacha terms. Import VAT would add a sixth, but the test pairs it with domestic VAT below and the pair does not clear. That bar is the Budget’s own, and it is low (section 5). A higher bar changes nothing: no line sits between Pay As You Earn’s 7.9 per cent and the royalty’s 13.4, and the Fund’s figures grow the economy by 12.8. Company tax, mining and non-mining together, is up 42.0 per cent on the 2026 Budget, and by August it had collected 89.5 per cent of that year’s budget. The VAT rate rise to 16.5 per cent from 16.0 covers 3.1 points of import VAT’s 5.0. Pay As You Earn’s 7.9 per cent comes before the new 39.5 per cent top band adds anything. The test prices the dollar lines at the 10 November fixing, not at the Budget’s rate. At that fixing, Table 3’s five dollar-linked revenue lines need 7.1 per cent more than their pace: ZMW99.1bn against ZMW92.5bn. At the Budget’s own rate the pace rises to ZMW97.57bn and they need 1.6 per cent.

Three lines depend on their new measures. Customs duty needs 14.0 per cent, and the Address raised the line ZMW1.6bn above the Green Paper’s figure while raising duties on ceramic tiles, crude edible oils and selected aluminium products (p.21). Excise needs 19.6 per cent even with fuel excise restored, against plastics excise raised to 200 per cent from 100 and a new 80 ngwee bottle duty. Export duties need 17.0 per cent on the new 5 per cent zinc concentrate duty.

The stretch is in four lines: user fees, withholding tax, domestic VAT and mineral royalty. They carry ZMW21.5bn of the ZMW30.7bn between the adjusted pace and the target. That is 2.1 per cent of GDP out of 3.1, a gap about the size of the planned deficit. None of the four depends on the fuel estimate, which enters only import VAT and excise. Withholding tax needs 21.4 per cent. The Estimates put ZMW18.2bn of the ZMW26.2bn line on rent, interest and similar payments, ZMW4.1bn above the 2026 Budget. Dividends carry ZMW2.6bn, ZMW0.8bn more, where the new 15 per cent on mining dividends applies (Address, para 119, p.22). Mineral royalty needs 13.4 per cent and is the least stretched of the four. The Green Paper assumes 1,083,122 tonnes at USD14,292 a tonne (its Table 1, p.6). The assumption is 16.5 per cent more output than the Address projects for 2026 (section 2), at a price 7.3 per cent above the January-to-August average; the target needs far less. The royalty steps up with the price (the scale in force since 2023). It takes 4 per cent of the first USD4,000 a tonne, 6.5 per cent of the next USD1,000, 8.5 of the next USD2,000 and 10 of the rest. The assumed price alone therefore lifts the royalty on each tonne by 9.4 per cent, to USD1,124 from USD1,027. At that price the line needs 3.6 per cent more output at the 10 November fixing, and no more output at the Budget’s 21.50. At that rate and price the target implies no more than about 911,000 tonnes of copper, below the 930,000 projected for 2026. The Address also cut growth to 4.9 per cent from the draft’s 6.0 and lowered this line only by its gold royalty measure. The risk here is the copper price and the kwacha, not output: if the assumed output arrives at the assumed price, copper alone pays ZMW2.8bn more than the target at the fixing.

User fees, fines and charges must rise 49.1 per cent above a pace that already credits 2026’s exceptional revenue. ZMW9,450m of the line is “Proceeds from Sale of Crops” under the Ministry of Agriculture (Head 89), nil in 2026. The same head spends the same amount on strategic food reserves, which the Estimates assign to outstanding obligations from the crop marketing season and the Food Reserve Agency’s running costs (pp.813, 820). The head’s output target for 2027 is 500,000 tonnes of maize purchases (p.820). For the sale, the Estimates give neither tonnage nor price. The Agency’s two-season maize bill runs near ZMW30.3bn (its board statement of 29 September 2026; the 2025 cost as reported to Parliament). On the count in our post of 6 October 2026, the volume it can export across a drought cycle could approach 3.1m tonnes. Recovering the bill on that volume needs ZMW9,774 a tonne, near USD500 at the end-September rate and USD480 at the 10 November fixing. At that price the proceeds need 0.97m tonnes and recover 31.2 per cent of the bill. The Agency offered maize near USD275 in September, with the regional market near USD230 (market reporting by ZAMACE, the commodity exchange, in “The Two-Sided Coin”, Canary Compass, 2 October 2026). At those prices and the fixing, the line needs 1.7m to 2.0m tonnes, more than the 1.6m tonnes the same post counted as exportable before the next harvest. Selling all 1.6m tonnes raises ZMW7.5bn to ZMW9.0bn. Without the maize the user-fees line needs 5.5 per cent, about what its ZMW1.3bn of new measures raise (Address annex, p.26). Take out 2026’s exceptional revenue as well, a one-off worth ZMW3,612m in the pace, and it needs 26.6. If none sold and the Treasury still paid, the deficit would be ZMW39.5bn, 3.9 per cent of GDP.

Domestic VAT must more than double after collecting 15.5 per cent of its 2026 budget in eight months. The Green Paper attributes the collapse to the slow uptake of smart invoicing (para 102). The line also nets refunds, and claims stood at ZMW60.3bn at end-September (Address, p.2). The Estimates cut the line to ZMW6.4bn from ZMW11.8bn and cap refund clearance at ZMW2.5bn a month (arrears annex). Importers reclaim VAT paid at the border against domestic VAT or as VAT refunds, so the test is fairer on the two lines together. Combined, they need at least 13.9 per cent, because the fuel adjustment counts fuel VAT before any refund. Grants, outside domestic revenue, must rise 71.9 per cent after partners delivered 41.6 per cent of the profile to August. ZMW3.5bn of the target is a new line the Estimates call “Various Donors - swaps”. Table 4 shows what else changed since the September draft.

The ECF explains the financing side: staff agreed the programme on 9 October, between the two documents, and the Budget carries its money. The Address books ZMW6,724m (USD312.8m in the Plan) of programme loans it labels IMF (p.20), a line the 2026 Budget did not carry. The Plan expects the Fund to approve the programme in the fourth quarter of 2026. The revenue side is mostly the crop sale; without it, domestic revenue is 22.35 per cent of GDP, the same as in the 2026 Budget to two decimals (Table 8 rounds that to 22.3). The new revenue measures together raise ZMW10.8bn, 4.6 per cent of domestic revenue (Address, para 122, p.22). On the spending side the Address cut the dedicated arrears line by ZMW10.3bn, while its paragraph 78 commits to prioritising arrears clearance. The ZMW4.7bn left is 9.1 per cent of the ZMW51.4bn of arrears on goods and services at end-June (Address, para 13). The Estimates leave the rest to liability management and offsets (arrears annex).

5. Is the Budget Realistic?

The Address never prints its GDP. Its spending of ZMW284.8bn at 28.4 per cent, and revenue and grants of ZMW242.9bn at 24.2 per cent, place 2027 GDP within 0.2 per cent of the Green Paper’s ZMW1,003.1bn (its Table 1, p.6). The Green Paper keeps 2026 GDP at ZMW923.7bn, the figure the 2026 Budget projected in September 2025, and grows it 8.6 per cent in nominal terms. At the Address’s 4.9 per cent real growth, that implies economy-wide prices, the GDP deflator, rising 3.5 per cent, well below the 6 to 8 per cent inflation band. The deflator is a wider measure than consumer prices, so the two need not match, but this gap is large.

The outturn since that projection has run faster. The Zambia Statistics Agency’s final 2025 GDP is ZMW797.7bn, so the Budget’s figure asks 2026 for nominal growth of 15.8 per cent. In the first half of 2026 nominal GDP grew 20.7 per cent on a year earlier, 7.4 per cent in volume and 12.5 per cent in prices, each from the Agency’s levels. In the second quarter it grew 13.6 per cent, with prices up 6.0 (The Monthly, Vol 282, September 2026, Tables 1 to 5). The quarterly figures sum to ZMW785.9bn for 2025, the Agency’s estimate by expenditure, against the final ZMW797.7bn. The year can still reach ZMW923.7bn. The strain is the step to 2027: prices rising 3.5 per cent after 12.5 in the first half and 6.0 in the second quarter. “The Two-Sided Coin” (Canary Compass, 2 October 2026) put 2026 GDP at ZMW902.7bn, the final 2025 figure grown at the IMF’s May 2026 projections. On that base the same 2027 level implies prices rising 5.9 per cent, just under the band’s floor. This Note quotes the Budget’s targets and Musokotwane’s projections on the Budget’s own GDP, and names the base wherever it uses another. If 2027 GDP proves higher than the Green Paper’s figure, the Budget’s ratios, the deficit included, turn out smaller than printed. Table 5 sets the Budget beside the Fund.

The two GDP bases in Table 5 differ by 7.5 per cent, so the kwacha figures compare better than the ratios. The Budget’s deficit is ZMW30.0bn and the Fund’s ZMW38.2bn: the Fund has ZMW4.8bn less revenue and ZMW3.4bn more spending. The Fund last updated its Zambia figures on 24 September 2026, before the Zambia Statistics Agency’s revision, the staff-level agreement and the Address; its 2025 GDP is still the preliminary ZMW725.6bn.

The White Paper on the 2026–2028 plan (January 2026), the approved version of last year’s cycle, shows how far 2027 has moved in ten months. It put 2027 GDP at ZMW1,046.6bn, which the Green Paper cut by 4.2 per cent. It projected growth of 6.5 per cent, a deficit of 1.6 per cent of GDP and net domestic borrowing of 1.9 per cent (its Tables 4 and 7). The Address plans 4.9, 3.0 and 2.9. In kwacha the deficit rose from ZMW16.25bn to ZMW30.0bn and net domestic borrowing from ZMW20.1bn to ZMW28.9bn. For 2027 the White Paper assumed USD8,983 a tonne and 1,206,410 tonnes; the Green Paper’s price is 59.1 per cent higher and its output 10.2 per cent lower (section 4). The higher price that lifted 2026’s mining lines is therefore already inside the 2027 base.

The Budget’s deficits here are cash deficits, the basis the Ministry reports (White Paper, January 2026, para 24); the 2027 figure includes the arrears paydown. Before ZMW55.7bn of interest, the 2027 Budget runs a primary surplus of ZMW25.7bn, 2.6 per cent of GDP. The Estimates say the domestic debt line, counted here as interest, also repays commercial loans (Head 99). The Address leaves little room for that: its target for net domestic borrowing, 2.9 per cent of GDP (p.3), is the full ZMW28.9bn its financing table prints (p.20). Repayments above about ZMW0.3bn inside the line would pull net borrowing below 2.85 per cent of GDP and change the printed 2.9. The primary surplus does not move with them, because repayments leave the deficit and interest by the same amount. Musokotwane’s 2026 projection leaves a primary surplus of about 0.4 per cent if interest ran to budget, against 4.2 planned.

On the Fund’s figures, gross public debt falls from 82.1 per cent of GDP in 2026 to 77.0 in 2027. The Address targets a 1.2 per cent deficit by 2029 (p.12) and the ECF a 3 per cent primary surplus by the same year (IMF Press Release 26/331). Met together, the two put interest at 4.2 per cent of GDP, against 5.6 in the 2027 Budget.

The gap between pace and target is ZMW30.73bn. Musokotwane’s measures raise a net ZMW10.8bn (section 4) and the crop sale ZMW9.45bn. That leaves ZMW10.48bn for growth to supply, 5.2 per cent of the pace. The pace includes ZMW3.6bn of exceptional revenue, 2026’s one-off scaled to a year; without it the figure is 7.1 per cent. Both are floors, because each assumes every measure delivers in full and all the maize sells. Both sit inside the 8.6 per cent by which the Budget grows the economy. At the Budget’s own rate of 21.50 the pace rises to ZMW208.0bn and the first figure falls to 2.6 per cent.

Without the maize sale, and with every measure delivered, revenue must grow 9.8 per cent above pace at the fixing, more than the 8.6 by which the Budget grows the economy in kwacha terms. The Budget is thin in three places: a maize sale with no tonnage or price, measures not yet enacted, and an exchange rate weaker than the market’s.

6. Who Buys the Paper

Foreign loans only just cover the foreign principal falling due (section 3), so a revenue miss, a stronger kwacha or a spending overrun lands on the domestic auction calendar unless the Treasury cuts elsewhere. The 2027 Annual Borrowing Plan, which the Address cites (p.16), sets gross domestic borrowing at ZMW115.0bn at cost: ZMW86.15bn to refinance maturing securities and ZMW28.85bn net.

Rollovers fund maturities at cost: the Treasury borrows again only what it first received. The discount, the gap between that cost and the face value repaid, is interest, as coupons are, and revenue pays both through the Budget’s interest line (Address, p.17). In 2031, ZMW29.8bn falls due at face: ZMW17.4bn is cost to roll over and ZMW12.4bn is discount to pay from revenue (Debt Statistical Bulletin, Quarter 2 2026 edition, Table 27). For 2026 the same table showed ZMW85.9bn due at face: ZMW66.2bn of cost to roll over and ZMW19.7bn of discount (End Quarter Four 2025 edition). A count at face overstates the rollover by that discount. The 2026 borrowing plan counts at cost: principal, with an estimate for bills sold and repaid within the year. The amendment presented to the House in May 2026, with the first supplementary, set gross domestic borrowing of ZMW108.6bn, with a net limit of ZMW29.12bn, leaving ZMW79.5bn of refinancing. The Quarter 2 2026 Bulletin, published later, supports that figure. It shows ZMW35.04bn repaid at cost in the first half and ZMW42.21bn due at cost in the second (its Tables 1 and 27). With ZMW2.11bn of bills sold since June and due by December (BoZ tender results), that is ZMW79.4bn.

For 2027, Table 6 builds the requirement from three pieces of principal at cost. They are paper outstanding at end-June (the same Bulletin, its Table 27; Chart 3), bills sold since, and the four bill auctions left this year at the amounts offered. We rebuilt the bills tender by tender from BoZ records, and the rebuild matches the Bulletin’s end-June bill stock to within ZMW1m. No bond sold since June falls due in 2027: the two- and three-year bonds were new lines and the reopened lines run seven years and longer. With the Budget’s net borrowing, the gross requirement reaches about ZMW100.1bn, 10.0 per cent of GDP, before bills issued and repaid within the year. The Plan’s ZMW115.0bn is ZMW14.9bn higher, all of it in maturities, and on our count bills sold and repaid inside the year fill most of it: ZMW13.7bn, their 2026 total (Table 6). The remaining ZMW1.2bn would be the last four 2026 bill auctions allotting ZMW2.5bn each at cost against the ZMW2.2bn offered. The Plan’s figure is 11.5 per cent of GDP, against the amendment’s ZMW108.6bn, 11.8 per cent of the 2026 Budget’s GDP.

Commercial banks held ZMW62.5bn of Treasury bills and bonds on 30 October, 48.2 per cent of their ZMW129.6bn of kwacha deposits (BoZ fortnightly statistics; Chart 4). The planned 2027 net borrowing equals 22.3 per cent of those deposits, before rollovers. The ratio measures scale only: banks will not take all of it, and their deposits will grow. The authorities told the Fund they would hold non-residents to 15 per cent of primary bond issuance from 2027, after 23 per cent in 2026 (IMF Country Report 26/21, January 2026). At two ZMW6.3bn bond auctions a quarter, the fourth quarter’s calendar, that admits about ZMW7.6bn a year. Non-residents’ net purchases in the first quarter of 2026 alone (section 3) came to about ZMW11.7bn at that quarter’s average rate, across both markets. The 15 per cent binds at auction only; bills and the secondary market stay open. The balance falls to banks, pension funds and insurers. At end-June pension funds held ZMW70.9bn of the paper at face value: 45.1 per cent of the bills and 21.8 per cent of the bonds (Debt Statistical Bulletin, Quarter 2 2026 edition, section 4.3).

Yields fell through 2026. The BoZ’s policy rate fell to 10.75 per cent in September 2026 from 14.25 in December 2025 (Address, para 11). The Treasury raised most of its money early: by August net issuance of securities had reached ZMW26.7bn, 91.7 per cent of the ZMW29.12bn limit for the year. The limit covers securities: the BoZ’s buyback loan (section 1) and ZMW2.8bn drawn from bank balances are separate lines (MEI, August 2026, Table 1.0). Buyers took more than the calendar offered: from January to 25 September the tenders allotted ZMW88.8bn at cost against ZMW79.6bn offered (BoZ tender results). On our reading, with most of the year’s money raised and little room left under the limit, the Treasury could turn away expensive bids. In September it met and then exceeded its offer, and still rejected 37.4 and then 54.5 per cent of the bids for its bills (Table 7). The 364-day yield fell from 11.9 per cent to 10.0 by 1 October, below the policy rate and 3.7 points above October inflation. The 15-year bond cleared at 15.45 per cent on 25 September.

The second supplementary resets the limit. Its ZMW18.0bn of additional domestic borrowing (as reported) lifts it to about ZMW47.1bn, against the ZMW26.7bn of net issuance by August: ZMW20.4bn of room against the ZMW27.3bn deficit of the last four months (section 1). The rest falls to foreign loans, net of repayments, or to cash. At its last two October tenders the Treasury took all but ZMW4m and ZMW9m of the bids, at an unchanged yield. The calendar still offers four bill auctions of ZMW2.2bn each at cost to 24 December, against ZMW9.5bn of bills falling due at face. It offers the quarter’s two bond auctions on 20 November and 18 December (BoZ public notice on the fourth-quarter calendar). The 20-year bond, the extension the Medium-Term Debt Management Strategy 2026–2028 prescribes (p.15), debuts at the first. What the Treasury must raise in 2027 depends on the cash it closes 2026 with, and neither the opening nor the closing balance of its funding account is public. The Budget’s net domestic borrowing is ZMW28.9bn, against ZMW21.4bn in the 2026 Budget.

On our reading, the risk to 2027 yields tilts up across the curve, most for bills. Yields have already fallen far: since January bills are down 350 to 450 basis points and bonds 269 to 334 (BoZ tender results to 29 October for bills and 25 September for bonds). Bills clear below the policy rate, and the Treasury now takes almost every bid (Table 7). The Plan’s net borrowing, 2.9 per cent of GDP, sits well below the 5.1 per cent of the 2026 Budget’s GDP that year’s limit reached (as reported). The lighter supply depends on the Budget holding. If no maize sells and the kwacha stays at the market rate, the deficit is ZMW43.8bn, 4.4 per cent of GDP. If the kwacha weakens from 20.38 to the Budget’s 21.50, a holder who counts in dollars loses 5.2 per cent on the exchange rate. If it holds at 20.38, the Treasury collects ZMW5.6bn less in revenue and grants (section 3). The 15-year bond’s September cut-off yield of 15.45 per cent stood 9.15 points above October’s 6.3 per cent inflation, the widest margin on the curve. The 20 November auction is its first pricing since the policy rate cut of 30 September.

Official Figures That Disagree

Official figures disagree in at least four places, three within the Address as delivered on 6 November and one between the debt documents; these four go to the Ministry as questions.

1. The corporate income tax rise. Paragraph 117 lifts rates by 5 points on income taxed at 20 per cent and below; annex item 24 covers all income taxed below the standard 30 per cent. The two cover different companies, so the company tax line depends on which the Finance Bill enacts.

2. The top Pay As You Earn band. Paragraph 118 starts it above ZMW37,000 a month; annex item 25 starts it above ZMW34,000. The ZRA’s statement of 10 November uses the higher figure.

3. The ceramic tile duty. Paragraph 111(a) of the Address raises customs duty on ceramic tiles “to 40 percent from 25 percent”; annex item 1 raises it “from 5 percent to 40 percent”.

4. External principal in dollars. The 2027 Annual Borrowing Plan repays USD550.52m in 2027, and the Debt Statistical Bulletin schedules USD658.03m (section 3). For 2026 that year’s Plan repays USD573.04m, and the Bulletin’s two editions schedule USD747.66m and USD678.16m (section 2).

Grading and Pre-Registration

Earlier work read the Budgets’ rates from other documents. “The Forecast Is Not the Evidence” (Canary Compass, 6 March 2026) used about 20 for 2026; our posts of 7 November used 21.97, and 19.49 for 2027. The borrowing plans convert at 26.00 and 21.50. For 2026 the direction held and the posts understated the size: the estimated January-to-August loss on dollar lines is ZMW20.6bn, where the posts printed ZMW8.9bn. For 2027 the direction reversed: the market is stronger than the Budget’s rate, not weaker. Four tests follow from this Note. First, Musokotwane’s projection of a 5.9 per cent deficit for 2026 grades when the 2026 outturn publishes. Second, this Note pre-registers the base question. ZMW923.7bn for 2026, ZMW1,003.1bn for 2027 and real growth of 4.9 per cent imply a price rise of 3.5 per cent, and the White Paper on the 2027–2029 plan will not keep it. If its GDP figures and real growth still imply 3.5 per cent or less, the call fails. Third, the first Note after the March 2027 MEI grades first-quarter collections line by line against Table 3. We expect user fees, withholding tax and domestic VAT each to collect less than a quarter of its target; the call fails on any line that collects a quarter or more. The call leaves the royalty out, because it turns on the copper price and the kwacha (section 4). Fourth, a post grades the 20 November debut against the 16.4 per cent extrapolated in “The Two-Sided Coin” (Table 10A).

Revenue Ratios Beside GDP Ratios

Of every kwacha of domestic revenue the 2027 Budget collects, 23.9 ngwee goes to interest, down from 28.4 in the 2026 Budget (Table 8).

Table 8’s external debt service ratio, 7.2 per cent, covers only the Budget’s own line. The Fund’s analysis, which counts non-residents’ kwacha paper as external debt, projects 13.4 per cent on average for 2026 to 2031, against its 14 per cent threshold (IMF Country Report 26/21). On 2026’s unadjusted collection pace, domestic revenue is 20.2 per cent of the 2026 Budget’s GDP, against the 22.3 budgeted and the 23.3 the 2027 Budget targets. On the Fund’s 2027 figures, gross public debt of 77.0 per cent of GDP is 3.0 times revenue and grants.

Parked Questions

Beyond the four disagreements, eight questions go to the Ministry of Finance and National Planning and its Debt Management Office. What cash balance opens 2027 in the Treasury funding account, which the 2027 Annual Borrowing Plan does not report? Does the non-resident limit at bond auctions stand? On what estimate of bills sold and repaid within the year do the Plan’s maturities rest: 2026’s total (section 6) or another? How many tonnes of maize, at what price and to which buyers, raise the crop sale proceeds, and what does the Treasury pay if they fall short? How much of the Domestic Debt Service line is principal on commercial loans? Why do the borrowing plans convert at 26.00 and 21.50? At 21.50 the royalty target implies no more than about 911,000 tonnes of copper (section 4): does the Budget stand on that output or on the Green Paper’s 1,083,122? What 2026 GDP and price assumption stand behind the Budget? One goes to the BoZ: how much of the supplementary borrowing lands in bills against bonds. One goes to the ZRA: the age profile of the VAT refund claims, and the clearance the 2027 VAT line assumes. And one stays with the House committees that reviewed the supplementaries: how much of 2026’s deficit is food security spending, Musokotwane’s second cause.

Close

The 2027 Budget’s real growth matches the Fund’s. With every measure delivered and the maize sold, its revenue target needs revenue to grow at least 5.2 per cent above pace at the 10 November fixing. Without the maize it needs 9.8, more than the 8.6 by which the Budget grows the economy in kwacha terms. Whichever way those and the exchange rate resolve, the difference falls to the domestic market unless the Treasury cuts spending. On our reading, the risk to 2027 yields therefore tilts up across the curve, most for bills. The 20 November debut, the Monetary Policy Committee of 25 and 26 November, the ECF Board date and the amended 2026 borrowing plan will confirm or move the numbers in this Note.

Sources

Ministry of Finance and National Planning (2027 Budget Address, 6 November 2026; 2027 Estimates of Revenue and Expenditure (Output-Based Budget), 6 November 2026; 2027–2029 Medium Term Budget Plan and 2027 Budget Green Paper, September 2026; 2026 Budget Speech, 26 September 2025; White Paper on the 2026–2028 Medium Term Budget Plan, January 2026; Monthly Economic Indicators, August 2026; 2026 Annual Borrowing Plan, September 2025, and its amendment as presented to the National Assembly in May 2026; 2027 Annual Borrowing Plan, November 2026; Debt Statistical Bulletin, Quarter 2 2025, End Quarter Four 2025 and Quarter 2 2026 (updated) editions, section 2.2 and Tables 1 to 27; Medium-Term Debt Management Strategy 2026–2028), National Assembly of Zambia (Supplementary Appropriation Act No. 78 of 2026; Supplementary Appropriation (2026) (No. 2) Bill, N.A.B. 79 of 2026, 23 October 2026; Mines and Minerals Development (Amendment) Act, 2022, for the copper royalty scale), Zambia Revenue Authority (2026 Budget symposium presentation, 29 September 2025; statement on the Pay As You Earn proposal, 10 November 2026), Zambia Statistics Agency (The Monthly, Vol 282, September 2026), Energy Regulation Board (monthly pump price build-ups, December 2025 to October 2026; price review, 30 September 2026; 2025 Statistical Bulletin), Food Reserve Agency (board statement, 29 September 2026), ZAMACE (market reporting, September 2026), Bank of Zambia (Fortnightly Statistics to 30 October 2026; interbank rates, 10 November 2026; Balance of Payments Statistics, Q2 2026; Monetary Policy Committee Statement, 30 September 2026; Treasury Bill Auction Results, tenders 01/2026 to 22/2026; GRZ Bond Auction Results, tenders 01/2026/BA to 07/2026/BA; public notice on the fourth-quarter 2026 auction calendar), IMF (World Economic Outlook, October 2026; Country Report 26/21; Press Release 26/331; debt sustainability framework for low-income countries, with the World Bank), Times of Zambia (23 October 2026) and News Diggers (4 November 2026) for Musokotwane’s statements to the House on the second supplementary, Canary Compass calculations.

Disclaimer

Canary Compass research is published by Canary Imara Ltd, a private company incorporated in Kenya (PVT-3B15OQP7). Canary Imara Ltd also trades proprietary capital in markets covered by this research and may hold positions in instruments discussed. Canary Imara Ltd and anyone who works on a Canary Compass piece do not trade an instrument it names between its final draft and its publication.

This publication is for information only. It is not investment advice or a recommendation to buy or sell any instrument; readers should seek independent advice before acting on it. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of Canary Compass and the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

Canary Compass publishes research on Africa’s monetary and fiscal architecture: Zambia and Kenya in depth, with our own forecasts, and the world, emerging markets and Africa read through the channels that reach them. This page sets out every plan, what it costs and what it includes.

Paid publishing starts soon. Pledge now and you pay nothing until it begins.

Free

• Every Friday Reflection.

• Series overviews.

• The opening of every paid piece, up to the paywall.

Paid: $25 a month or $250 a year

• At least 40 Notes a year on Zambia, Kenya, Africa and the global economy: four in every publishing month.

• Quick Takes on major events, within 24 hours on weekdays.

• Our own forecasts for Zambia and Kenya, scored in public: a short forecast table in our monthly Zambia and Kenya Notes, and from March 2027 a full table three times a year for each country, graded against what actually happened.

• Our flagship archive.

Founding member: $500 a year

Everything in the paid plan, plus:

• A quarterly call with Dean, recorded.

• One question a month, answered within two business days.

• The numbers behind every chart we publish.

• A say in what we research next.

• If you want, your name on our supporters’ page.

Founding membership is limited to 50 and closes when paid publishing starts.

Institutional seat: $1,500 a year

Everything in the paid plan for one named reader, plus:

• Five answers a month.

• The quarterly call.

• The numbers behind every chart we publish.

Billed once a year by invoice.

Desk licence: $5,000 a year

Everything in the paid plan for up to five named readers, plus:

• Ten answers a month.

• The quarterly call.

• The numbers behind every chart we publish.

• A watermarked PDF of each paid piece when it publishes.

• The compiled PDF of each series.

The names can change each quarter. Billed once a year by invoice.

Private briefing: $2,500 each

• One private briefing for your team, including new analysis.

• We agree the scope in writing first, and you pay when you book.

Training and advisory

• Courses and bespoke work, quoted for each engagement.

How the extras work

• Answers: each answer runs to 300 words, draws on our published work and arrives within two business days. A question that needs new analysis becomes a private briefing.

• The quarterly call: 75 minutes, from the world and emerging markets to Africa, Kenya and Zambia, with questions sent in advance. It is recorded with chapter marks, so you can replay only the part you need. The first call runs in April 2027, after our first full forecast tables. No call runs in July or December.

• Chart data: the values behind each published chart, each number with its source and period.

Good to know

• Canary Compass publishes ten months a year. Nothing publishes in July or December. The annual plan pays for those ten months. The monthly plan bills every month.

• These are web prices, at canarycompass.com. In the Substack iPhone app, prices are higher to cover Apple’s fee, and founding membership is available only on the web.

• If our prices rise, monthly, annual and founding subscribers keep the price they joined at for as long as they stay subscribed. Seat and licence prices are set each January.

• You can cancel any time and keep access until the end of the period you have paid for.

• Substack emails a receipt for every payment, which you can use for expense claims.

• Every plan receives each piece at the moment it publishes. No plan gets early access.

• Institutional seats, Desk licences, briefings, training and advisory: dean@canarycompass.com

Canary Compass will move to paid subscriptions soon. Founding members put a question to Dean every month and join a quarterly call. Founding membership is limited to 50 and closes at launch. Pledge now; you pay nothing until paid publishing begins. Friday Reflections will stay free. See every plan and what it includes. Institutional licences: dean@canarycompass.com

This week, about 32,000 retirees in Paraguay learnt that they will collect their November pensions at other banks, most of them assigned by their pension institute, the Instituto de Previsión Social. Payments go out from 17 November. On 3 November the central bank put their old bank, Ueno, into resolution, its solvency below half the required minimum. The institute had about USD345m of its members’ money there in deposits and bonds, some 9 per cent of its portfolio by its own count. Ueno paid about 11 per cent, the second-best rate in the market. A report in July had found the institute over its legal limit for deposits in a single bank, which it said it then corrected. In late October members and workers’ groups asked it to pull the money out. Its investment director replied that the bank met the institute’s risk criteria, and that term deposits were not easy to withdraw. A week later the money sat trapped. The central bank’s own president had run Ueno’s predecessor before taking office in 2023.

It was Paraguay’s second pension story of the year. In March the country reformed the Caja Fiscal, the pension fund for its teachers, judges, police and other public servants. In the sectors running deficits, workers’ contributions rose from 16 to 19 per cent of pay, and for the first time the law wrote in a state contribution, set at 10 per cent. Teachers, university staff and judges now need to be at least 53, with 25 years of contributions, before they retire. Teachers protested outside Congress as it voted. A ruling-party deputy called the reform barely a patch: by her estimate it cuts the deficit from about USD800m to about USD600m.

One country, in one year, has raised every question a pension member should ask. Does my money earn a proper return, safely, towards a pension that can actually be paid? Can I choose how much risk it takes, or move it when I am worried? Can I use part of it before I retire, without wrecking my retirement? And does the state pay what it owes into the pot, and keep its hands off the investment decisions? Paraguay has just forced every one of them into the open in a single year. Anyone with an African pension should put the same four questions to his own.

Until 1997, Canada did with its national pension what many African governments do today. Workers’ contributions to the Canada Pension Plan were lent straight to the federal and provincial governments, in 20-year bonds the fund could not sell. Then Parliament changed the law. It set up an independent board and gave it one job, owed to the plan’s members. In the law’s words, that job is “to invest its assets with a view to achieving a maximum rate of return, without undue risk of loss.” It also made the law hard to change. An amendment needs the federal government and two thirds of the provinces, holding two thirds of the population. The lock has faced two tests. In March 2024 about a hundred business leaders asked the finance minister to push pension money into Canadian companies. The fund replied in public that pension funds “are not designed to achieve regional economic development or corporate subsidies.” In October 2025 a minister pressed again, and the fund held its line. By March 2026 it held 793.3 billion Canadian dollars (CAD), about USD570bn, and had earned 8.8 per cent a year over ten years.

Canada is the opposite case. Its fund answers to one mandate, and has defended it against its own ministers in public. It answers the first and the last of the four questions superbly. It barely touches the middle two, and for one pooled plan covering some 22 million Canadians, that is a sound choice. Members who want choice and access need a different design.

Pension schemes come in two basic shapes, and the shape decides which questions a member can even ask. In the first, the pension is a promise. The law fixes what he will receive, usually from his pay and years of work, and the scheme and the state behind it must find the money. Specialists call it defined benefit. In the second, the pension is an account in his own name. Contributions go in, investment results add or subtract, and at retirement he gets what the account has become. That is defined contribution. In the public promise schemes of this story, the state carries the final bill. So it guards the promise on paper and is tempted to skip today’s payment, because the hole lands on a future government. In an account scheme the member carries the result, so he has every reason to want a say. Paraguay’s Caja Fiscal is a promise, which is why its deficit fell on the state to close, and why the March reform reached for contributions and the retirement age.

Paraguay’s members asked, and their institute said no. In America’s 401(k), the account shape built for the member, the money answers to him. Each worker’s payslip feeds his account, and his employer usually tops it up. He picks investments from a menu, and if he never chooses, many plans sit him in a fund that grows safer as retirement nears. He can borrow from himself where his plan allows it, up to half of what has vested and never more than USD50,000, and repay through his payslip.Malaysia shows that a national fund can pay well and still open a door. Its Employees Provident Fund (EPF) has paid dividends above 5 per cent for three years running, on a legal floor of 2.5 per cent for conventional savings. Since May 2024 each contribution has been split three ways. Three quarters goes to retirement, 15 per cent to housing, education and health, and 10 per cent to a flexible account the member can draw on at any time. The retirement share stays locked, and Malaysia learnt why the hard way. During the pandemic the government let members withdraw 145 billion Malaysian ringgit (MYR), about USD34bn, in four rounds. Afterwards more than half of its members under 55 had less than MYR10,000, about USD2,200, left for old age.

The mechanism behind Paraguay’s March law is simple. When the money coming in cannot cover the pensions promised, a government has three ways out within the pension itself. It can collect more, by raising contributions. It can pay less, by cutting what each pension is worth. Or it can pay later, by raising the age at which pensions start. Paraguay used two of them at once. Italy shows where that road runs over the years.

Italy’s state pension is a promise paid from today’s money: today’s workers fund today’s pensioners, with no large pot behind them. Italy spent the equivalent of 15.3 per cent of GDP on pensions in 2024, among the highest shares in Europe. In November 2011, with the eurozone in crisis, Italy’s borrowing costs rose more than 5.5 percentage points above Germany’s. Berlusconi resigned, and on 4 December 2011 Elsa Fornero, the new labour minister, presented the pension reform. She broke down at the press conference and could not finish the word for sacrifices. Women’s retirement age rose towards 66, and the reform moved every worker’s years from 2012 onto a pension worked out from what he had paid in. Today the standard age is 67, and from 2027 it rises again. Italy used all three ways out, and the one people remember is the age.

Africa’s version of the crisis has a name: arrears. The state, as an employer, owes the pension funds money it has not paid in. Sometimes it took the worker’s share from his payslip and never sent it on. Sometimes it skipped its own share. My reading is that one of the three ways out is already in use. Italy wrote paying later into law, one rule for everyone. Much of Africa practises it one retiree at a time, as a bill left unpaid. Kenya’s National Social Security Fund (NSSF) is the account shape, by its 2013 Act. Every member has an individual account, and his pension is whatever that account can buy when he retires. The Act calls the government a guarantor of the public interest in the fund; it guarantees no pension. Civil servants under 45 moved to a contributory scheme of the same kind in January 2021. So the Kenyan member carries his own result without the account’s freedoms. His employer may move part of the contributions to a private scheme; he himself may move nothing. Inside the fund there is one portfolio, and a 25-year-old and a 59-year-old sit in it together. His one door is narrow, a pledge of part of his benefits to back a housing loan where his scheme allows it, and the courts killed the wider one, withdrawal for a home, in 2022.

The Kenyan state also falls behind on what it owes. Its counties owed pension schemes 115.9 billion Kenyan shillings (KES), about USD900m, at the end of 2025, by the Parliamentary Budget Office’s count. The bill is unpaid deductions, the interest and penalties on them, and the funding hole they left behind. At the national level, a public servant leaving service waits 195 days, on average, for his benefits; on 5 October the President ordered them paid within ten. Retirees of Nairobi’s water company have waited close to 20 years.

Nigeria chose the account shape in 2004 and then built most of its freedoms. Every covered worker has a Retirement Savings Account with a licensed private administrator. Since 2018 the money has sat in funds sorted by age, and a worker under 50 may move one step bolder. He can move to another administrator once a year if he loses faith in his own, the step Paraguay’s members could not take. Since 2022 he may put up to a quarter of his balance towards the equity a mortgage lender requires on a home. A worker out of a job for four months may draw as much as a quarter of what he has saved. The Nigerian state leans hard on that pot, and owes it money too. More than half of pension assets sit in federal government securities, and the regulator has said that concentration limits the system’s ability to beat inflation. In February 2025 the government approved a bond of 758 billion Nigerian naira (NGN), about USD505m, to clear its pension debts, including its own unpaid employer contributions for 2017 to 2021. By mid-2026 the regulator reported that backlog paid. The states owed another NGN626.8bn, about USD435m, in 2025, by BudgIT’s count.

Zambia keeps the promise shape for its National Pension Scheme Authority (NAPSA), with an actuary checking whether the fund can pay what it owes. In April 2023 members got their first door, a one-time withdrawal of a fifth of their contributions. A new National Pension Scheme Act, signed in June and in force since August, writes the doors into the main statute. A member may take a fifth of his indexed contributions early and another lump sum just before retirement, with a smaller minimum pension in return. The same Act widens the state’s side too, letting the fund enter joint ventures and co-investments.

Zambia’s civil servants sit in an older promise with almost nothing behind it. The Auditor-General found that the Public Service Pensions Fund holds assets covering 9 per cent of what it owes, at its last actuarial check. Like Paraguay’s Caja Fiscal, it is a promise the state must fund.

Ghana runs both shapes, by tier. The first, run by the Social Security and National Insurance Trust (SSNIT), is a promise. The second, worth 5 per cent of pay, is an account run by licensed private trustees. Ghana’s 2008 law lets a member pledge his second-tier benefits and use a voluntary third-tier account to secure a mortgage. In 2022 the regulator called those provisions largely unused. The door was in the law, and almost nobody had walked through it. Ghana also shows how a state can reach into the pot on terms, and pay its way out. Its 2022 debt exchange exempted every pension fund after a strike threat. The state came back in July 2023 with a voluntary offer for the funds alone. Face value stayed whole, the bonds swapped into new ones due 2027 and 2028, and a side bond kept most of the old interest flowing. It was a request rather than an order. The state had also left more than 3 billion Ghanaian cedis (GHS), about USD290m, unpaid to SSNIT over a decade, by a World Bank count. Then it caught up, reporting the arrears cleared by 2025 and itself current since.

South Africa runs both shapes side by side. The Government Employees Pension Fund (GEPF) is a promise, guaranteed in law, and it held more than it owed at its last valuation. A commission of inquiry into the Public Investment Corporation, which manages most of its money, found in 2020 a dysfunctional board and deals done without proper care. When the governing party floated forcing pension funds into prescribed state assets, the country argued it in the open and dropped it. In September 2024 South Africa also opened a wide member door. Under its two-pot system, a third of every new contribution goes into a savings pot the member can draw on once a tax year, while two thirds stay locked. It covers the government fund too. The arrears there sit with thousands of defaulting employers, by the regulator’s count: private firms, municipalities, and a state-owned Post Office. Its workers’ fund has seen 12 cents for each rand of contributions never paid in.

Run the five against the four questions and a pattern shows. On returns, only Nigeria makes its administrators publish their returns each month, where anyone can mark them. On choice, Nigeria alone writes risk-sorted funds into every account by law. On access, South Africa, Nigeria and Zambia have working doors, Ghana’s stays on paper, and Kenya’s is one narrow pledge. On the last question, whether the state pays what it owes and keeps its hands off, every one of the five has fallen short on one half of it or the other. Ghana has paid its way back, and Nigeria’s federal bond has now done the same; the counties and states behind them have not. Of the five, Kenya’s members carry the hardest combination. Its members bear their own results, like Nigeria’s, with little of what Nigeria gives back.

Three things follow. The first is that arrears and infrastructure are the same problem. Governments ask pension funds for long-term money for roads and power while failing to pay what they already owe in. What Kenya’s counties owe their workers’ schemes is nearly four times the up to KES30bn, about USD230m, NSSF has put forward for the Rironi to Mau Summit toll road. They are different funds, but it is the same pool of Kenyan pension savings. Arrears are a loan nobody approved: taken from workers without a vote, past the due dates the law sets, at a penalty rate nobody collects. Every unpaid deduction is capital the pool never received. The second is that the member’s door is cheap. The usual objection to 401(k)-style access is that withdrawals drain the long money a continent needs. The record so far says the drain is small. South Africa’s two-pot withdrawals came to under 1 per cent of its retirement industry’s assets in their first five months. Malaysia’s members took about 1.4 per cent of their fund through the flexible account in two years. Both doors open onto new contributions and a capped sum seeded from the old balance, while the bulk built over decades stays locked. That is the design. Chile is the warning on the other side: three pandemic rounds, a tenth at a time, paid out about USD50bn by early 2022 and left three million accounts at zero.

The third is that Africa’s pools are small in part because so few people are in them. Kenya counts 21.6 million people in recorded work, and only about 3.5 million of them in modern-sector jobs. A trader or a motorbike-taxi rider will rarely lock money away in a pot he can never touch, whatever it earns. So a door, held to a slice, need not shrink an African pension pool. It may be how the pool grows, because it gives the people outside a reason to come in.

That leaves the real tension, between the member who wants his money working for him and a continent that needs patient capital for large projects. It can be settled by design, and the parts already exist in the countries above. Where the money is actually saved, split the member’s pot into a locked core and a smaller slice he can use, as Malaysia and South Africa do. A promise scheme prices any door against the pension it reduces, as Zambia’s new Act does. Hold cash for the pensions and draws the fund already expects, so a draw does not force a sale. Let the core carry the nation’s long projects through priced vehicles, under ceilings a trustee chooses to fill, never floors a minister orders filled; South Africa argued about exactly that and chose the ceiling. Cap what the fund may place with any single bank or company, which is the guard Paraguay’s institute breached. And make each fund publish, every January, the return it commits to beat that year, with inflation the test beneath it, then report against it monthly, judged over several years. A target written down before the year is the same discipline I described in Marked to Market: the condition set before the position, with a copy in other hands. Australia goes further: its regulator tests the main retirement products against benchmarks built for their own asset mix, and after two straight failures the regulator closes the product to new members. None of it works if the state keeps borrowing from the pot through arrears. So the first move is to make arrears hard to create rather than easy to punish. The pension deduction should leave with the salary, in the same payment, straight to the scheme; for Kenya’s counties, the scheme’s share would come off the national transfer before a county ever receives it. South Africa’s Treasury worked the same pipe from the other end in July 2026, freezing transfers to municipalities that had pocketed workers’ deductions, Johannesburg among them, until the municipalities settle the pension debts. This has a real cost. Counties fall behind because cash is short and priorities drift, and paying at source moves that shortfall into the open, into unpaid suppliers or thinner services. That is also the point. The shortfall stops hiding inside workers’ pensions.

The state will always be in the pension pot. It writes the law, appoints the biggest funds’ trustees and borrows from them all. The state is a tool here, not a saviour, and what it owes first is what it already owes: the contributions it took and never paid in. Pay those, give the member his slice, lend the nation the core at a price, and let a law, not a minister, hold the line between them. On 17 November, Paraguay’s retirees will be paid at banks most of them did not choose, from a fund they could not move. Ask your own pension the four questions before your November payslip does.

A companion note this month sets out the full design: the ratios, the table and the order to build it in.

A revision to last month’s reflection, Your Pension Financed a Losing Trade. Nobody Asked You. At publication, the reported prices for the Kenya Pipeline Company block ranged from KES9.02 to KES9.44, and I wrote that the proposed dividend went with the shares to the Dangote group. Reporting since then, citing people close to the deal, gives fuller terms. The selling schemes asked for KES9.50 a share and the buyer refused. He paid KES9.04 a share in cash and gave up the KES0.32 dividend on the block, so the sellers keep it. The sellers therefore receive about KES9.36 a share. The buyer’s headline of about KES9.44 counts the dividend he gave up, and the remaining KES0.08 may be dealing costs, which nothing published settles. On the block, the sellers’ gain comes to about KES1.6bn, about USD12m, against about KES2.4bn, about USD19m, that a one-year Treasury bill would have earned over the same months. The verdict stands. Reports named NSSF and the public service scheme among the sellers. The share price has since risen well above KES9, closing at KES11.60 on 5 November. That rewards whoever kept their shares, and nobody deciding in February could have known it was coming without inside information. None of these terms has come from the regulator or from any scheme.

AI-illustration: Pension money in motion, February to October 2026.

Update, 6 November 2026: a correction to this reflection runs in The Lesson From Paraguay for Anyone With an African Pension.

Kenya spent this week arguing about pension money. The argument followed a sale. In February, state-linked pension schemes and funds were reportedly pushed into buying Kenya Pipeline Company. In October they sold a block of it. The conversation took me back. Three years ago this month I published an article called A Collaborative Reflection on Bridging Zambia’s Prosperity Divide: Wealth Creation, Alpha, Literacy and Fund Management. I read it again while the argument ran, and I counted six questions in it, all about the money workers put aside for old age. Kenya has now answered them, without meaning to.

First, the public facts. In January and February the government sold 65 per cent of Kenya Pipeline Company at 9 Kenyan shillings (KES) a share. Rich private investors stayed away. Foreign investors from outside East Africa ended up with about 0.02 per cent of the whole company. Retail buyers ended up with about 2.5 shares in every 100 of it. The sale was heading for failure. Business Daily later reported that the State pressured cash-rich parastatals and state-backed funds to save it, and that funds were selling bonds that February to raise the cash. NSSF, the national social security fund, alone invested about KES36bn, a stake valued at about KES38bn in later filings with the regulator. In all, thirteen state-linked schemes and agencies, plus Uganda’s national oil company, were reported to have supplied KES95.8bn of the KES106bn needed.

Why did so few Kenyans buy? Nobody can poll the buyers who never came, so I will not claim their reason. What can be shown is what they had been told, and what they did next. Research houses published their sums in January and February. The sceptics’ fair values ran from about KES3 to about KES6 a share, against an asking price of KES9. One bank called it a premium price for a mature, regulated company. Others noted that the dividend on offer could not compete with tax-free government infrastructure bonds. The research was public, and the private money stayed home. The allocation results show it. The offer was open to private buyers, local and foreign. Together they took under 5 shares in every 100 on offer. The largest Kenyan buyers who did go in were the ones reported to have been pushed.

There was a second barrier, older than the price. Being allowed to buy is not the same as being able to buy. I wrote about this in November 2023, when Kenya was lining up sales of state companies, and I listed what a sale built for citizens would need. Set aside part of the offer for low-income buyers, and price that part lower so they can reach it. Help them pay, through government and partner programmes. Sell through the groups people already trust, such as cooperatives and village banks. Ask the new owners to hold for ten years. Keep majority ownership of key companies at home, by law, for the life of the company. Then report in public, at regular intervals, who ended up owning what. The pipeline sale had a retail window, but none of the rest. It offered no lower price for the poor and no financing built for them, and it promised no special report on final owners. So Kenyans stayed out, whether by judgement or by exclusion. Then they were brought in anyway, through their pension money, without being asked.

Nearly eight months later, on 19 October, about a quarter of the company changed hands in a single deal, going to the Dangote group, the Nigerian industrial giant. The sellers were reported to be several of the Kenyan state-backed holders. The block was roughly 4.5 billion shares, which had cost KES40.5bn at the February price. The reported prices lay between KES9.02 and KES9.44, and the reports differ on where in that range the block went. Five days before the sale, the company’s board had proposed its first dividend as a listed company, KES0.32 a share, which shareholders still have to approve. They sold before the date that decides who gets paid, so the entitlement went with the shares to the buyer.

Then came the opinions, in three camps. The first camp celebrated: the funds bought at KES9 and sold above it, so the trade made money, and a win is a win. The second camp said nothing was lost. State funds bought a state company, and the state banked the proceeds. Money just moved from one government pocket to another. The third camp, led by a legislator’s thread, said this was all a plan for a takeover through a future rights issue, a sale of new shares to existing owners. That is speculation, and I will not adopt it. There was also a fourth, quieter strand. Some voices did ask what the trade meant for pensioners and for the independence of their funds. They were not the ones setting the week’s terms.

The three camps disagree on almost everything, yet they share one blind spot. None of them looks at the deal from the seat of the people who own the money. The second camp treats pension funds as a state pocket. The third treats them as pieces in someone else’s game; even its charge that the funds were made to sell cheap is about who moved the pieces, not what the owners were owed. The first camp cheers a headline profit for the funds without asking what the owners gave up to get it. Put the owners back in the room, and none of the three answers the question that matters. Did the members get value?

A pension scheme holds money in trust for its members. Calling a fund a state scheme tells you who wrote the law that set it up, or, in the civil service schemes, who employs the members. In NSSF’s case, most members work in the private sector. Either way, it does not tell you who owns the money. Contributions are wages, paid later instead of now. Money paid in has changed hands; the employer has no more claim on it than on last month’s salary. The government can regulate the fund and appoint the people who guard it. It can even lean on it. None of that makes it the owner. The owners are the clerks, teachers, drivers and nurses whose payslips built these funds. Very little of the week’s noise was addressed to them.

The second camp fails first. One pocket to another only works if both pockets are on the same coat, and these are not. The buying pocket belonged to workers. The receiving pocket belonged to the state. Some will say the same citizen stands behind both, as worker and as taxpayer. Even then, the two pockets follow different rules. One is held in trust for him, with his name on it. The other is spent for everyone. And the trade between them was not a wash. The workers’ pocket gave up cash and got shares most other buyers had refused at that price. When the state banked KES106bn, much of it was money that had left several million private owners, at a price of KES9, with no owner asked. Saying the government came out fine answers nothing. The question is whether the members came out fine.

The first camp fails on arithmetic. A return means nothing on its own. You must compare it with what the same money could have earned over the same months, in a choice the holder really had. Usually you have to guess at that choice. Not here. Some funds had just sold bonds to put cash into this offer. The honest benchmark is the lower-risk government paper they gave up. Nobody has published exactly which paper was sold. The dealer spoke of bonds, and that February the tax-free infrastructure bonds were paying into double digits, above the one-year bill. So the one-year bill is a gentle benchmark, used the simple way: buy it in February, count eight months of interest. Rates fell through the year, so a bill sold at month eight would have fetched a little more than that straight count. And the bill sits below the paper the funds reportedly sold. Kenya also sells three-month and six-month bills, but neither reaches across the eight months on its own, and rolling them would mean guessing the rates ahead. Only the one-year bill covered the whole window at a rate known on the day the choice was made.

So compare, on that KES40.5bn block. One-year government paper yielded about 9 per cent a year in February. Held from the offer’s close in late February to the sale on 19 October, it would have earned about 6 per cent of that KES40.5bn, around KES2.4bn. The sale’s reported numbers tell three competing stories, and all three fall short. Business Daily reported most of the block at KES9.02; on that story the gain was nearer KES90m. The week’s loudest defenders put the sellers’ profit at KES1.6bn, which works out to a KES9.36 price. And Dangote was quoted at about 327 million US dollars, which at around 130 to the dollar is the KES42.5bn put on the block, about KES9.44 a share, a gain of about KES2bn before costs. The stories cannot all be literally true, because a block averaging KES9.44 cannot also have printed mostly at KES9.02. One reading nearly reconciles them. Suppose the buyer’s total includes dealing costs of between about half and one and a half per cent, the top being the legislator’s own estimate. The implied share price then works back to between about KES9.30 and KES9.40, close to the defenders’ KES9.36, and the gain to between about KES1.4bn and KES1.8bn. That reading challenges the KES9.02 report, and nothing published settles how the fees were treated. The verdict survives every version: the biggest story, KES2bn before costs, trails the bill’s KES2.4bn, and the reconciled readings trail it by over half a billion. The proposed dividend changes none of this. A buyer paid one price for the share and the KES0.32 entitlement together, so the sellers were paid for both, and the comparison with the paper already counts everything they received. The sellers themselves are known only from newspaper reports; the regulator has said nothing. No scheme has stated in public what it sold, what it kept, what price it got, or what the sale cost.

Whatever anyone intended, and nobody has shown a plan, the record shows the job the money did. For eight months the pot carried the stake from a state that wanted out to a buyer who came later, and it bore the risk of the wait. A warehouse charges for that. This one was never paid a fee.

Were the February sceptics wrong, then? Money could do three things with February’s offer: stay out, go in and hold, or go in and sell. Eight months have sorted all three, at this week’s prices. The saver who stayed out and kept government paper earned the bill’s roughly 6 per cent. The holder who bought in February and never sold has done best so far. Three days after the block, the share touched KES9.78, lifted by Dangote’s arrival, and he sits on an uncashed gain of almost 9 per cent. The schemes did worst, and theirs was the only money whose owners were never asked: in at KES9, out before the rise their own sale set off, and below the bill at every reported price. And one actor had no February seat at all: the buyer, who paid the sellers’ price in October and watched the market mark his block up within days. So the market has moved against the sceptics’ prices; the block went through above the offer, not at KES3 or KES6. But hold the dates apart. The holder’s gain exists because Dangote came, and nobody deciding in February could have known he was coming without inside information. A decision is judged on what could be known when it was made, and the research February could read pointed away from the offer. The lift came from a strategic buyer paying for influence over what the company becomes, something no small saver was ever offered, and a block that size rarely fetches the screen price. On the test that mattered, the sceptics’ method was right, and even the money that refused out-earned the forced trade. The asset can come good, and the trade can still have left the members worse off than the paper their schemes sold. The saver’s refusal was not fear of the unknown. It was arithmetic.

One buyer at this table behaved like an owner. Uganda’s national oil company took 20.15 per cent of the whole company, paying about USD255m. Before its money moved, it negotiated two board seats and a veto over hiring and firing the chief executive. Kenya’s schemes put in more money and took no such terms. NSSF’s chief executive was given a seat on the board months after listing; Uganda’s terms were signed before its cash left home. The offer was the same and the price was the same. One investor charged for its money. The rest handed theirs over.

The week argued about the money and the price. Control came up only as a plot, never as a question. A pipeline is the artery that moves a country’s fuel. In February, Kenya sold 65 per cent of it. By October, a quarter had moved to a foreign group. And Uganda, through the veto it negotiated in February, has more say over who runs Kenya’s fuel line than the Kenyan workers whose savings paid for the rescue.

The pipeline was not even this year’s biggest sale. In June the same government sold 15 per cent of Safaricom, the country’s most valuable company, to Vodacom for about KES204bn, raising the Vodafone family’s stake to 55 per cent and handing it control. In September the High Court ruled that sale unconstitutional, partly for want of real public participation, and ordered the shares returned. Appeals are running and the shares are in dispute while they are heard. But sit with the court’s first reason. The public was never properly asked. Each sale carried the same two failures: foreign hands came out stronger, and the owners were not asked. Safaricom passed into foreign control outright. The fuel line kept its Kenyan majority, but its strongest levers, the largest private stake in Nigerian hands and a veto in Kampala, sit abroad. Twice in one year the state sold, and the explaining was left to courts and newspapers. There are good reasons for a government to sell shares; citizens can own them instead, and that is the point of listing. Selling what is strategic, while the rest of the world guards its own, is a different act, and nobody has offered a reason for it. The pipeline sale was even financed by the savings of the country’s own workers. Nobody asked them either.

Zambia shows both halves of this lesson. In 2021 it brought the Mopani copper mines back into domestic hands, taking them over from Glencore. The idea was right, because the prize was real. Copper is the metal of electrification. Demand for it was already set to run when Zambia moved, and the supply troubles since have only added to the upside. A country that owns its copper owns a claim on that future. But the vehicle was wrong. The buyer, the state mining company ZCCM-IH, was far smaller than the deal it signed. The deal left Mopani owing USD1.5bn, more than 30 billion Zambian kwacha (ZMW) at 2021 exchange rates, with ZCCM-IH guaranteeing it. Going into the deal, at the end of 2020, the buyer’s entire equity stood near ZMW16bn, so it guaranteed roughly twice its own worth. Zambia’s pension fund, NAPSA, held about ZMW55bn at the end of 2021, more than three times that equity. Even for NAPSA, Mopani was too big for one bite.

Three years later the present administration brought in a group from Abu Dhabi, which took 51 per cent, mostly by putting new money into the mine. Control drifted foreign again. To be fair, the mine needed money and mining skill, and the deal brought both, and this time the money went into the asset itself. But the fights over metals and supply chains were already running in 2021, so the case for holding on was visible even then. In hindsight, two better routes may have been open. NAPSA could have built its stake in steps. Or the foreign money could have come in with a sunset written into the deal, so that ownership folds back home over time. If the Abu Dhabi deal carries such a clause, I have not seen it published. Bringing strategic assets home takes vehicles made to carry them. A sovereign fund built for the purpose does that job: it buys and holds what matters, with money put aside for it. Patient pension money can sit alongside such a fund, on one condition. At a price and on the trustees’ own terms, in slices their balance sheets can safely carry. That is investment. A forced march in at KES9 is conscription. And letting strategic assets go because the treasury is having a bad year trades fifty years of control for one year of cash. Zambia’s dilution at least fed the mine. Kenya’s sale money sits in new state funds, with the spending still a promise.

Underneath this week sits an old argument, and it is not a Kenyan one. Every country with a pension pool faces the same temptation, because development is always urgent and the pool is always there. May the state sacrifice the returns of tomorrow’s retirees for the nation it wants to build today? The defence is the same everywhere: the project matters, and retirement money is the deepest pool of long-term savings in the country. Both things are true, and neither one answers the question. I am not arguing that pension money must never build the nation. It can. But the nation must pay for it like any other borrower. The returns must fit the risk, and over time they must beat what the money was already earning and keep ahead of rising prices. That is all a benchmark is: the check that the pot was kept safe.

And watch what the pension money actually funded here. Not the pipeline. The company pays for its own works out of earnings and borrowing, and the buyer now talks of expansion money of his own. The members’ money funded none of that. It bought the state out of its shares, and the proceeds went to the state’s new funds, where the building is still a plan. Whatever builds the pipeline will be money that charged for coming. The only money in this story that set no price for coming was the pensioners’.

Priced development money is not a fantasy. Zambia showed one version of it in 2024. NAPSA, Zambia’s national pension scheme, committed USD300m towards the new Lusaka to Ndola road. By its own account, it lent in dollars, at 9.5 per cent a year, for 13 years, standing first in line for repayment. Toll money sits in a locked account to pay the loan back. NAPSA also says nobody directed it; the deal was brought to it, and it said yes on its own tests. None of this makes the road a good project, and a coupon does not prove the price was right for the risk. Those answers will take years. And the difference between Lusaka and Nairobi runs deeper than debt against shares. Uganda bought shares in the same pipeline and still charged for it. The real difference is whether the money set any terms at all. NAPSA set terms. Kenya’s schemes were handed a price.

So what should this pool be doing instead? Not sitting in government paper, waiting for the next summons. Some of that paper even carries an infrastructure label, tax-free, and the schemes buy plenty of it. But a label is not a project. The money goes into the state’s hands and mixes with everything else, and whether every shilling reaches infrastructure is a question the label does not answer. Real project lending looks like NAPSA’s road: a named project, a price, security, and tolls set aside for repayment. A pool this deep could do more of that, on terms.

The week also showed this pot living under two standards. When the state wanted the pool, the pot was treated as open: no price and no consent, only pressure. When a member wants his own balance, the pot becomes sacred. Protection rules apply, with one narrow door for housing. Today the law lets members of any registered scheme pledge part of their benefits to guarantee a housing loan, where their scheme has set that route up. A second route, drawing savings out to buy a home, was opened in 2020; the High Court quashed it in 2022. Beyond the house, nothing. Your savings can be marched into a state company, but they cannot stand behind your shop or your working loan. Many countries lock members out to protect the pot, and that alone is no scandal. The scandal is locking the member out while letting the state in. Both rules cannot be right at once. If protection is the principle, it must bind the state first, because the state is the biggest hand reaching into the pot. And if the pot can take risk after all, the member’s own future has first claim on it. Let him borrow against a slice of his balance, at a fair price, with a floor no loan can touch, so a failed shop cannot eat a whole retirement. Banks already refuse to count much of your record when you ask for credit. I wrote about that last month, in Cronyism Without a Crony, the lock with no villain to fire. This is the same citizen at the next door: his record cannot speak for him, and now his savings cannot either. Today, outside your own house, your pension is not allowed to say a word.

When other states have reached into the pot, they have at least done it in daylight. When Argentina took over private pension funds in 2008, even that seizure was done through a law passed in congress, which citizens could read and fight. When South Africa argues about directing pension money, it argues in parliament, in the open. Kenya passed no law. The pressure was quiet, and we learned of it from a newspaper. Kenya’s pension law already demands independent judgement from trustees. Quiet pressure is how that judgement gets displaced without anyone changing a word of the law. And however the pressure arrives, by statute or by phone call, the thing being spent is the same. A pensioner’s retirement is not the state’s to spend.

A document is missing from this whole year. The Kenyan funds and agencies alone put more than KES60bn into this sale, and money like that does not move without paper. Somewhere there are trustee resolutions, committee memos, due diligence reports, and minutes that record who voted for this and why. In eight months of national argument, not one page has surfaced. No trustee of any selling fund has explained the entry or the exit. No regulator has demanded an explanation in public. Members should be shown the amounts, the dates, the costs, the alternatives weighed, and who approved it. Kenya spent a week arguing about an investment case that nobody who made it will show us.

This is not a one-off. In September I published a note on Kenya’s sports bond, the borrowing behind the new Talanta stadium, called Converting Vices into a Virtuous Sporting Cycle. It found two state schemes holding 54 per cent of a single bond issue. I asked why then. Nobody has answered.

Savers in this region have already said what they want. In 2023 I was part of a survey of middle-class savers in five African countries, published as Unlocking African Prosperity. At the time I set out its findings in A Financial Paradigm Shift for Africa. Overall, 73 per cent said they put protecting their savings ahead of chasing high returns. The Kenyan share was the same. That may sound like an argument against benchmarks. It is the opposite. The same survey asked about rising prices. In Kenya, 57 per cent wanted growth that clearly beats inflation, and only 14 per cent would accept growth below it. Zimbabweans, who have lived through money dying, wanted growth above inflation most of all, 7 in 10 of them. Protecting the pot means protecting what the money can buy. A pot that grows slower than prices is losing value quietly. The benchmark is how you check. Other systems hand the member a dial for exactly this. In Britain, a member whose workplace pension builds a pot in his own name can choose how that pot is invested, cautious or bold. In Kenya’s NSSF and Zambia’s NAPSA, and in most statutory schemes like them, no such dial exists. One portfolio is picked for everyone, and the member’s appetite for risk is never asked. On every story that can be checked, the pipeline trade, as the sellers ran it, took more risk than the government paper some schemes sold, and earned less. The owners had told anyone who surveyed them that they wanted the opposite. The system never gave them a way to say it where it counts.

The 2023 article was about Zambia, not Kenya. The subject first found me there. Zambia sets a hard test for any pension pot. In the ten years to 2023, inflation there averaged about 12 per cent a year, and the kwacha kept falling against the dollar. A pot had to grow at that pace, on average, just to stand still. Zambians know this. In the same survey, about half of them asked only for growth that keeps pace with rising prices. Of the five countries surveyed, they asked for the least.

I have made the mistake I am describing, though I had help. That article opened on my own statutory pension, with the same NAPSA that priced the road. A pre-retirement withdrawal forced me to run numbers I had never run. My own sums at the time put the return over my contribution years below inflation. Its buying power had not been protected; it had shrunk. A scheme can set terms on one road and still fail the member’s test over the years. For years I contributed without once asking for the comparison I am now demanding from Kenya’s trustees. Yet asking had a wall in front of it. My contribution statement showed what I had paid in, and nothing else: no interest or indexation, and no sign of what the money had become. A member cannot question numbers he is never shown. The scheme followed its rules. Nobody had asked whether the rules were enough. Retirement is the wrong day to find that out, and for many members it is the first time anyone checks.

So back to the six questions from 2023. I asked them of Zambia. Kenya answered them this week. They belong to any country where workers set money aside. The first said pension returns must be measured against outside markers, inflation and the currency above all, and beat them over time. This week a trade was cheered with no marker in sight. The second asked whether trustees, stopped today, could account for every position and what it earns. No selling trustee has yet accounted for this one. The third called for compulsory training and certification for everyone guarding collective money. Kenya requires trustee training today; the question the week leaves open is what that training is producing. The fourth said fund managers must be judged on the return they add, with consequences when they subtract. This year brought praise and accusation, but no consequence. The fifth put financial literacy at the top of the national list. The cheering for the smaller number tells you where that stands.

The sixth question was not for trustees or regulators. It asked for members who engage and who question. That question is for you, because you own the money. The law does not ask members to vote on investments; it asks trustees to use their own judgement for you, as if you were watching. Nobody asked you whether your deferred wages should rescue a listing, and nobody had to. That is exactly why the watching matters. Remember what your savings are allowed to do for the state, and what they are not allowed to do for you. A trustee who has never been questioned by a member will, in time, guard the money like nobody is coming back for it. Ask your scheme what it bought in February and sold in October. Ask what the money was earning before it moved. Members are not mute. They grumble to each other about statements that show them nothing, and some long ago stopped asking because nothing ever answered. But a grumble lands on nobody in particular; a question put to the scheme sits on someone’s desk. And a question like this has to be taught before it can be asked, so those of us who know it owe the teaching.

AI-illustration: the stack at desk scale, and the cap that stays on your side.

This week Canary Compass published four essays asking who owns the artificial intelligence stack. One subject kept coming up around them: ask one of the Chinese frontier models about matters that touch China and it may simply decline. The behaviour has been documented since DeepSeek’s breakout launch early last year, and I have tested it myself on different occasions: the returns deflect or decline. A tool that declines whole subjects its owner is sensitive about carries that owner’s boundaries into your work. The instrument has an address.

The American machines are not clean either. Daily use teaches you each one’s temperament. One is lazy until pressed. Another hedges until the answer says nothing. A third pads its reasoning with aphorisms and filler, and exaggerates what it has found. The same one wants you pleased: it flatters, and it will force a connection that makes no sense rather than tell you the pieces do not fit. A fourth hands you unrelated output with full confidence. More than one will offer you, in perfect prose, a story about yourself that never happened. Those are craft defects; every workshop has them, and every machine has fences of its own. A decline that tracks the owner’s politics is a different failure. The temperament is the instrument’s. The boundary is the owner’s.

The series closed on its answer yesterday: not Africa, at any floor where global terms are set. At the desk the question is different, and it is not whether to use these machines. You have already started. The question is what you handed over when you did.

These machines cross filings, transcripts and languages in minutes that a desk would need weeks to cross. Look at the arrangement with a trader’s eye. The model is trained by someone else, on terms someone else set, behind boundaries someone else drew, at a price someone else can change. And what you type into a hosted machine lands on its owner’s servers, which is why many institutions will not let their best material anywhere near one. The one asset that is unambiguously yours is the thinking. It is also the easiest to hand over, because handing it over feels like leverage.

In June, in a conversation among friends, one of us was preparing a trade around a coming listing and had taken the idea to a frontier model. He told the group the machine was winning the argument. I gave him one sentence: do not give the machine the thinking cap. Keep your position. Then set the machine on it as the most resourceful critic you have ever hired, and let it attack with everything it can reach. Move only on arguments you understand, never on its say-so; a fluent objection can be a wrong one. A position that survives has earned its next look, and the trade stays yours to make. A position that dies has cost you pride, not capital. Either way, the thinking stayed yours.

Kenya sorted me on the difference between the two postures early. I only understood that later. I grew up under the 8-4-4 system in the eighties and nineties, and that system was heavy. From about grade three to grade eight, children of eight to fourteen carried heavy bags to school, because we did ten subjects. The list: Mathematics, English, Kiswahili, Science and Agriculture, GHC for Geography, History and Civics, CRE for Christian Religious Education, Art and Craft, Music, Home Science and Business Education. Picture the class-three version of yourself: shorts, tiny body, big head, and a bag that looks like it is carrying the child rather than the other way round. And it was competitive. Schools were ranked nationally, and you waited by the television to see the top schools and the top students read out. Under that pressure the incentive structure for teachers shifted from teaching students to making them cram. My last year of primary school was a year of prep exams. At the national examination the ten collapsed into seven grades of 100 marks each, and the one number out of 700 usually determined where you ended up.

Then came my own secondary school. Excellent schools selected on that examination, and the roll served them well. Strathmore did not. It ran its own timed test, and it had usually picked its new Form One class before the national results were even out. As far as I can remember, the test ran to something like 30 questions in five to seven minutes, and nobody was expected to finish. The questions were the school’s own, so you did not know what was coming, and there was nothing to prepare from beyond what school had already built in you. The score was how many you got right inside that window: how you reasoned under the clock, not how much of the syllabus you had carried in. I did not know what it was measuring then. I do now. Psychometricians call that design a speeded test, the architecture the Wonderlic made famous: more questions than the clock can hold. Cramming had nothing to grip. The school was selecting for the cap.

The cap has a voice inside the labs now. Tom Zahavy, a DeepMind researcher, argued in January this year, in a position paper of his own titled “LLMs Can’t Jump”, that the machines have mastered induction and are advancing in deduction. What they still lack, he argues, is the third way humans reason: the unprompted jump to a frame the data never contained. The claim is contested, and the June advice does not hang on it. On his division, criticism is work the machine can do; the jump to a new frame is the move it cannot yet make. Whichever way that argument resolves, the frame you choose and the conclusion you sign stay yours to answer for.

The stakes are higher where I write from. Frontier machines are paid for in dollars here, out of local incomes. What the reach costs sits on the invoice. The judgement about what the reach returned is the job, and the job does not transfer. Keep your own hands. Borrow the better vision.

The borrowed vision has a habit. Too often these machines skim where the work needs thoroughness. Hand one a heavy paper expecting everything back and you may be shortchanged: a reading that covers the surface and misses the load-bearing middle. The habit cuts across every platform, and a subscription does not cure it. An expert catches the gap. The reader in danger is the one who came to be taught, who cannot see what is missing or check the summary against anything, and the machine does not say what it skipped.

What does an institution do on Monday morning? Keep the judgement on payroll: every machine output that matters carries a named person who holds the position on it. Set the machine as critic, never as author: it attacks the house view, and the forming of the view stays human work. And open a refusal ledger. When a tool declines a task, write down the date, the tool and its version, the task, and the reason it gave. A few months of that file says more about your dependence than any vendor briefing. It is a record of where your instruments’ owners end and you begin.

The week asked who owns the stack. Friday’s question was always yours to settle: who does the thinking. These machines are extraordinary. The one that declines teaches more about who owns them than any essay could. Keep the cap, for as long as it is yours to keep.

AI-illustration: the stack on the scale, the floor rising beneath the pan.

AI Week, Part 4 of 4. Part 3, published last night, priced the lines and ran the scenarios. This part runs the market test: the translation, the funding circuit, fourteen objections and the answer, with the series' consolidated Sources. Friday's Reflection closes the week. A 20 minute read. Data cut-off: 8 October 2026.

In brief

• The index’s AI earnings now run through the semiconductor complex, and the market prices the crest: Nvidia trades below the index’s own forward multiple. The valuation holds only if the hyperscalers’ capex starts returning, and Goldman has the investment swinging from tailwind to drag by 2028.

• Followed to its tenants, the buildout wagers on rent, and the tenant whose budget never asks what the compute returned is the state, whose terminal use case this continent already hosts. The funding circuit is machinery under construction: bill-heavy issuance into a bid being built captive, with a quiet tax on nominal holders near the size of the hole a displacement scenario opens.

• Fourteen objections run against the series’ position in the open, and the strongest survive as scenarios rather than rebuttals. On the evidence assembled across four parts, the title’s question resolves against Africa at every floor where global terms are set, and the treasury Part 1 predicted is arriving.

9. The Market Translation

Part 3 assembled the rescue’s premise. This part runs the market test. For the reader who holds the index rather than the argument, the structure the series has mapped prices as follows.

The S&P 500’s earnings engine rotated through 2026. Semiconductors were set to supply roughly 44 per cent of the index’s second-quarter earnings gains on LSEG’s July estimate, a leadership transfer JPMorgan’s Chipenomics note flagged in May. Goldman Sachs attributes almost half of 2026’s index earnings growth to AI-related investment (September 2026). The market already prices the crest. Nvidia traded near 17 times forward earnings in late September, its cheapest since 2015 and below the index’s 19 (as reported, 29 September 2026): the market pays less for the engine than for the train. To hold the index’s multiple, the baton must pass back to the hyperscalers, which requires the capex Part 1 sized to start returning. Morgan Stanley’s July unit-economics frameworks price 25 to 50 per cent returns on invested capital across three GenAI business models, highest where the developer owns the infrastructure (Morgan Stanley Research, July 2026). The deadline is also dated: Goldman expects AI investment to swing from earnings tailwind to drag by 2028, as capex growth slows and depreciation climbs. From there the multiple rides on returns rather than on spending, and if the returns do not arrive, the earnings story and the multiple go together. The bull case needs durable profit beyond the investment cycle, in the owners’ rents or their customers’ productivity, and the index is priced for the owners.

The scale comparison makes the same point from outside. Goldman’s September estimate puts hyperscaler capital spending near USD 800 billion in 2026, a broader set than the four builders whose USD 720 to 745 billion Part 1 recorded. It is projected to reach about USD 1.2 trillion in 2027 and USD 1.4 trillion in 2028 (Goldman Sachs, September 2026). Enthusiasts call that the largest technological revolution in modern history. It is also among the largest concentrated capital commitments ever assembled outside wartime. Both descriptions carry the same balance sheet, and only one of them prices the downside.

Part 3’s five scenarios are therefore also multiple regimes: expansion, the state’s floor under a correction, unassisted compression, rupture taking the multiple off the table, and diffusion moving the rents off the model layer. The market is priced for the first, while the BIS warnings, the contested consolidation judgements and the off-balance-sheet plumbing Part 1 traced suggest the distribution carries fatter tails than the multiple admits.

The commoditisation question has to be answered in revenue terms, because the valuations depend on the answer. If the model layer is where margin goes to die, the frontier labs must earn somewhere else, and the candidates are few. They can move up into applications, where consumer subscriptions and enterprise software carry real revenue against real competition. They can move down into infrastructure, renting the compute every model needs whether or not the model itself is free. They can move sideways into sovereign and defence contracting, which is revenue that never appears in a commercial comparison. Or they find scale in none of the three, which is the compression branch arriving by another route.

Follow the capital and the answer is mostly the second. The buildout is a wager on rent, not on model exclusivity, and rent survives commoditisation, because a landlord does not mind that the tenants’ furniture became cheaper. Today the landlords are the hyperscalers and the infrastructure owners; the labs mostly rent. That reading joins the answer the rehearsal in Part 2 gave, where two successive waves of cheap Chinese models failed to move capital expenditure at all. Part 2 read the hold as a sovereign floor beneath the market; the rent wager is what the floor stands under. It also explains why the leading labs look less like software companies each quarter and more like utilities with research divisions attached.

The wager carries its own risk. Rent requires tenants, and if deployed demand arrives at a fraction of the forecast, the infrastructure strands while the model layer that was meant to justify it has already commoditised: the third scenario’s compression. What escalates it to the fourth is the rescue the stranding would summon, because the issuance is the trigger Part 3 placed in the absorption.

The geography of the rent is the part Africa should read closely. Both coalitions are exporting infrastructure rather than ownership. One offers to build independent AI infrastructure on partner soil, in the cable pitch Part 3 quoted: “They build it. It’s yours.” The other offers the application cooperation centres and training seats Part 2 counted. Neither transfers a floor of the stack. Both make the recipient a tenant of a rent-generating asset sited on its own territory, which is the mineral economy’s arrangement transposed to compute.

Stress the rent thesis at its weakest joint, which is the tenant. The frontier labs are the largest tenants, and they are also the layer commoditisation erodes, which makes their capacity to pay a function of the same trend the rent was meant to survive. Enterprise tenancy is real and its returns disappoint while budgets still grow, the Bain reading Part 1 carried, and the named-firm caps it recorded are the leading edge. Self-hosting on cheaper hardware subtracts from premium demand rather than adding to it. Strip those three and one tenant remains whose budget does not answer to a return on investment.

That inference belongs in the scenario column rather than the finding column, and the evidence keeps pointing at it. Sovereign funds are already deploying at scale, and the Genesis Mission has put the federal scientific estate behind acceleration. The procurement receipts are current. The Pentagon’s AI office signed contracts of up to USD 200 million each with four frontier developers across June and July 2025. In May 2026 eight companies, OpenAI and the largest clouds among them, agreed to deploy their models inside the Defense Department’s classified networks (Nextgov, May 2026). The lab carrying the Pentagon’s national-security-risk designation Part 2 recorded was left off the list. The Genesis Mission’s first funding solicitation is USD 293 million for AI-driven science (Department of Energy, March 2026). State demand large enough to underwrite a rent economy is concentrated demand: defence, intelligence, and monitoring, alongside science and administration, because the security uses are where a government buys compute without asking what it returned. An architecture whose economics depend on state tenancy has its principal use case chosen for it. And the state buys from either stack without return discipline, which is how Part 2’s forced choice reaches the rent economy.

Africa already hosts the completed version of that sentence. The surveillance layer Part 1 counted across at least 16 countries is no forecast here. A narrower, costed mapping prices at least USD 2 billion spent across the 11 countries it studied, the supply overwhelmingly Chinese (Institute of Development Studies, March 2026). If the rent thesis resolves toward state tenancy, Africa has no need to wait for a preview. The reference implementation is already running.

One objection disciplines that claim and sharpens it. Mass surveillance has never required frontier compute. Cameras, network taps, and metadata collection run on modest hardware, and the systems already installed across the continent were built well before this buildout began. Were monitoring the driver, the capital would look cheaper and far more distributed than it does. What frontier capability changes is comprehension, not collection. Surveillance states have always drowned in their own material, holding far more than they could ever read, which is what made collection tolerable. The archive outgrew its readers. A system that can summarise, cross-reference, and rank a population’s traffic removes that constraint, and the appetite for it has no natural ceiling, because comprehension improves with every model generation. That appetite is why the state is the tenant of last resort rather than one customer among many.

The listing calendar completes the translation. Both leading closed labs filed confidential draft S-1s in June 2026, Anthropic on 1 June and OpenAI a week later (company statements, June 2026). OpenAI’s chief executive ruled a 2026 listing out in September, retiring the within-the-year reports Part 2 carried (Fortune, September 2026). Read against the baton logic, the timing is structural rather than incidental: the model layer enters the public pipeline at the moment the index needs a new carrier of AI earnings growth. The layer with the deepest losses in the stack is being readied for the public at the moment the stack most needs external capital. That is the financing loop’s exit ramp, and it deserves to be watched as one.

One further asymmetry sits in the market data. Independent estimates put Chinese frontier models at roughly a third of American operating cost, with measured performance close to parity on general benchmarks (Alpine Macro, July 2026). Those readings are benchmark-dependent, and government cyber evaluations from the same month placed the leading Chinese model well behind. The estimates also bracket a wide range: a threefold cost gap at the frontier, against the hundredfold task-cost spreads Part 1 recorded between the American top tier and the cheapest Chinese models. What has not converged is the valuation. Private marks for the American labs run at multiples of their Chinese counterparts, and that gap widened as the capability gap narrowed. Part of the difference is capital access rather than capability, since firms shut out of Western exit markets are priced by a smaller pool of bidders. Which is the same argument arriving from the opposite direction. Value settles where the structure permits it to settle, not where the capability is.

The state’s tenancy has to be funded, and the funding answers the question Part 3 left at its close: who absorbs the fall. Start with why the spending becomes compulsory rather than discretionary. The federal revenue base leans on the thing displacement erodes. Individual income and payroll taxes carried about 85 per cent of US federal receipts in the first seven months of fiscal 2026 (Congressional Budget Office, April Monthly Budget Review, May 2026). Corporate income tax, the levy on what AI creates, carried under 6 per cent and fell 23 per cent on the year. Run a displacement scenario through that base, with the assumptions stated. USD 1 trillion of displaced wages, roughly 6 per cent of US compensation, removes about USD 270 billion of tax at an assumed combined payroll and income take near 27 per cent. Replacing 70 per cent of the lost income adds about USD 700 billion of transfers; the replacement rate is the scenario’s policy assumption, not a programme on the books. The first-order hole, before the transfers’ own taxability and feedbacks, is close to USD 1 trillion a year. Displacement digs it twice: it creates the transfer bill and removes the revenue base in the same motion. It is a scenario rather than a reading of current data; the null result Part 3 carried still stands.

The funding circuit for holes of that size is largely assembled, and it was not assembled for AI. Treasury issuance has shifted toward bills, 22.2 per cent of marketable debt at the end of July 2026, above the 15 to 20 per cent band the Treasury’s own advisory committee recommends. Bills were set to absorb an implied USD 409 billion of the September quarter’s USD 739 billion borrowing (US Treasury refunding and advisory committee materials, August 2026). The front end is being fitted with a manufactured buyer. The GENIUS Act of 2025 lists Treasury bills among the approved reserves for dollar stablecoins. Issuers already hold nearly USD 200 billion of bills and short-dated Treasuries, more than Saudi Arabia’s holding, inside a market near USD 300 billion (New York Times and Treasury remarks, as reported, 2026). The Treasury Secretary has called a market past USD 2 trillion reasonable (Senate testimony, as reported, June 2025). The long end is managed rather than defended. In August the Treasury announced it would at least double its liquidity-support buyback caps in the 10 to 30 year sectors as long yields rose (US Treasury and Bloomberg, August 2026).

The quiet tax closes the loop. Debt held by the public stands near USD 32 trillion. Each percentage point of inflation above the yield paid transfers roughly USD 320 billion a year of real value across the full stock, and nearer USD 230 billion on the fixed-rate stock alone. The bills that reprice within weeks are the price of keeping the bid. Bill-heavy issuance into that bid, rolled rather than repaid and serviced by nominal growth, finances a rescue without courting a new creditor; the appropriation is the part that still needs a vote. At three points the tax matches the scenario’s hole in size on the full stock, about two thirds of it on the fixed, and only one of the two required legislation. The erosion retires real debt; the cheques are written by the issuance.

This machinery extends Scenario four rather than contradicting it. Part 3 put Scenario four’s trigger in the absorption, confidence snapping on the supply the rescue implies. The circuit names the precise failure. The supply breaks nothing while the captive bid holds, so the break arrives if the captive bid walks, and the fences being built around the exits are what to watch. The machinery also answers who absorbs the fall. The tax lands on whoever holds the nominal claims, the ones fixed in dollar terms: savers, pension funds, money funds, and every balance sheet that clears in dollars. Part 3 priced the African arrival through the rate channel rather than the creditor base, and the circuit is that channel’s machinery, running on everyone’s money at once.

The buildout’s own paper has joined the competition for the same savings. Goldman tracked nearly USD 500 billion of AI-related debt issuance across markets by early August (Goldman Sachs Exchanges, August 2026). AI-related issuers accounted for roughly 18 per cent of investment-grade supply and 40 per cent of issuance at maturities of 15 years and longer. Debt is set to fund about a third of hyperscaler capex this year and next. Before any rescue is priced, the channel Part 3 named is already carrying weight: every borrower that is not the stack, African sovereigns included, now prices against the stack’s bid for duration.

The African seat in this trade is the uncomfortable one. African pension and sovereign funds holding US equities are long the very distribution described above. The continent priced out of the product is long the product’s equity risk. The tenant also holds the landlord’s shares, and a compression scenario lands on the same balance sheets that already underwrite the buildout through Treasuries.

10. The Stress Test

Fourteen objections, engaged in the open.

Africa is not literally absent from the middle floors. Correct, and the essay’s test was never absence. African-owned data-centre, connectivity, and application businesses exist at regional scale, and Africa Data Centres is the cleanest rebuttal to any absolute reading (Cassava Technologies disclosures, 2026). The test is system-level control: no African firm sets the global terms of frontier silicon, hyperscale cloud, frontier-model access, or the capital structure behind them. Regional ownership matters and should compound. It does not yet move the stack’s ownership map.

There are not literally two vertically integrated suppliers. Correct. The Western stack is a multinational production network, and an African institution can technically mix American cloud, Taiwanese fabrication, Korean memory, Chinese weights, and Gulf capital. The two-stack claim is geopolitical: two dominant strategic ecosystems around which controls, procurement, capital, and standards are organised. Hybridisation stays possible until conditionality prices particular combinations out, and that conditionality, not technological purity, is the mechanism the trajectory documents.

Concentration may not last. Open-weight challengers exist and improve; the regulatory moat is contested. Accepted in part: concentration stands here as the current structure, not a permanent one. But the tether argument cuts against the strongest version: if the challengers’ economics are substantially distillation, their independence is borrowed, and the retrieval and infrastructure layers concentrate regardless of what happens at the model layer.

The demand bottleneck may be overstated. Every general-purpose technology created demand categories its forecasters could not model: the internet produced the app economy and the gig economy from nothing. Static models have lost to dynamic reality every prior cycle. Held open. Part 3 prices both outcomes, and in neither does control of the indispensable layers move.

The bailout is likely, not certain. The 2008 rescue was politically radioactive, and AI companies hold no household deposits. Conceded: Scenario two is a base case, not a law. The Genesis commitment, the advisory alignment, and the pre-assembled sovereign floor move the probability, but the political economy is unsettled, which is why the essay carries five scenarios rather than one.

Technology diffuses. Smartphones went from luxury to universal in 15 years, and this cycle should commoditise the same way. Partially accepted: the model layer is commoditising in front of us. The answer is Part 1’s reach argument. The model may democratise. The retrieval layer, the verification capacity, and the infrastructure do not, and the open-weight wave relocates the paywall Part 1 priced rather than removing it.

The mineral leverage may be less than claimed. Correct, and the claim here is the smaller version: construction leverage, not performance leverage. The distinction is load-bearing precisely because overselling the endowment is how the negotiating window gets wasted on the wrong asks.

The sovereign floor may be too small. Genesis opened with a USD 293 million solicitation against hyperscaler capex near USD 800 billion. Correct on magnitude, answered on function: Genesis prices the intent, the reserve-currency balance sheet Part 3 named prices the capacity, and a floor is a backstop, not replacement demand. Nor must a rescue preserve the equity: 2008 floored the system while wiping or diluting the shareholders, and a state that buys assets in bankruptcy floors the infrastructure without flooring the marks.

The consensus serves its narrators. PIMCO positions for the repricing it warns of. Citadel sits on the frontier side of the bifurcation it names. The BIS case argues for wider supervisory scrutiny. Amodei’s coalition doubles as a moat. All true, and the July escalation sharpened the symmetry: dystopia warnings from labs preparing public listings, capture accusations from an adviser positioned in their competitors, a President’s bench split between them (WSJ, 21 July 2026). The state belongs in the same ledger. On 22 July the White House science office alleged that K3 was industrially distilled from an American model, and the Treasury Secretary put sanctions on the table. On 23 July the Commerce Secretary announced that K3 remained behind American frontier models, and a senior adviser declared the panic over. Both messages can hold at once, and each licenses a different policy. Alarm justifies enforcement. Reassurance justifies light-touch regulation. The state is no neutral referee between the labs but a third narrator running a dual mandate, and that week it ran both messages inside 24 hours. The same test applies at the poles: the loudest dismissals of the capex panic come from portfolios long the loop’s tightest node. This essay uses their data and discounts every framing that pays its framer, its own included: the author writes from the continent whose position the essay argues is mispriced.

Orchestration beats the frontier. Fused panels of cheap models match frontier benchmarks at a fraction of the price. Engaged and rebutted by Part 1’s reach argument. Benchmarks measure reasoning with identical inputs. Workflows measure reasoning across unequal reach plus the judgement to verify the return. A higher benchmark score closes neither the reach gap nor the verification burden, and the skill tax on the cheap path lands on exactly the institutions least resourced to pay it.

The self-tax objection. The sharpest version came from the conversations among friends that run through this series, in four steps. Entity-list Chinese models and you force American companies onto expensive American intelligence. The rest of the world keeps using cheaper models. Distillation does not stop because Washington published a list. The net effect taxes the competitiveness of the entire country to protect domestic labs. Conceded on the cost: the tax is real, and the state will pay it, as it paid the Huawei tax, because security floors outrank efficiency at the sovereign level. Answered on enforcement: restriction was never going to operate at the model level. It operates at the bundle level, through the silicon chokepoint every model trains on and the dollar system every economy clears through, with the precedent already priced in the trajectory above. The strongest version of the objection adds history: industries that competed from behind trade barriers have tended to atrophy behind them, from steel to Detroit, and protection can dull the champion it shields. Conceded, and noted for what it cuts: that risk falls on the protector, not on the forecast. And closed on the destination: the objection’s two failure branches, a shrunken gated American stack or accelerated forced alignment, both terminate in the world this essay maps. The mess is the route there.

The free-rider objection. The strongest African form of the diffusion argument, put to the author directly in the same conversations: we are not Americans, so why care who captures the frontier economics? Cheap capable models serve African users regardless of provenance. Whether Moonshot built or distilled, the capability arrived, and consumers care about what the tool does, not how it was made. Conceded on the near term: cheaper models are an immediate gain, conceded from the start. The rebuttal is a balance-sheet argument. Africa’s exposure to the surplus economy, the bloc that exports its industrial overcapacity, already runs three layers deep. The layers are industrial, per The Forced Choice; financial, per the 2026 Inflection series; physical, through the telecom and surveillance infrastructure Parts 1 and 2 document. Adding the intelligence layer to the same counterparty does not diversify a dependency. It completes one. And the continent has run this experiment before, at the goods layer. The subsidised import wave that made everything cheap also helped price Africa’s own manufacturing out of its own markets, and the thin industrial base the continent carries today is the settled ledger of that bargain. Cheap was the mechanism, not the benefit. A generation later the same offer arrives at the intelligence layer, from the same counterparty, at the same discount. The warning here is the continent quoting its own receipts. Layer by layer, the dependence converges on a single counterparty, and one holding a country’s industry, debt, networks, and intelligence layer holds its policy autonomy as the residual. Benchmark parity is flattening; market structure follows the spend. The consumer question and the sovereign question have different answers, and states that answer only the first discover the second has been answered for them.

The series promised a buyer and produced a backstop. The sharpest form of the prediction objection: four contracts capped at USD 200 million and a USD 293 million solicitation cannot make the treasury the marginal buyer of a build near USD 800 billion. Conceded on the magnitudes, and Part 3 conceded it first: “Disclosed sovereign procurement remains far smaller than hyperscaler capital expenditure, which is why Scenario two is a forecast and not a description.” What the receipts price is direction. The state is assembling the machinery of tenancy: procurement at the classified edge, a scientific estate put under solicitation, and a funding machine that needs no new creditor. Arriving names that direction, and absorption stays where Part 3 put it, in the forecast column.

The funding circuit fails its own stress test. First, the hole is a choice. The corporate share of receipts is partly the 2025 reconciliation act’s expensing provisions at work, and the customs line was rewritten in the same window. A state that can tariff imports can levy the AI surplus directly. Conceded, and the concession feeds the thesis: the political economy that assembles the rescue is the one that shields the surplus from taxation. Second, the captive bid is not captive: stablecoins redeem at par on demand, the most flight-prone creditor class in the market. Part of the bid substitutes for deposits and money funds that already held bills, and the inflation tax is the event that would send it walking. Third, the tax shrinks on inspection: net of the repricing bills and the indexed stock, three points transfer nearer USD 700 billion than a trillion in a year. Fourth, the circuit is circular in the shape this series faulted in private form, because a compression impairs the claims that fund the tenant of last resort at the moment the tenancy is needed. All four are conceded as the price of a machine still being built. The bid is being assembled by legislation and rails rather than volunteered, and the gap between USD 700 billion and the trillion measures the distance left. The loops fail differently, the private one commercially and the state one politically, which is where Part 3’s Scenario four keeps custody of the question.

11. Synthesis

Strip the essay to its findings and five sentences remain.

The intelligence economy is one chain from Zambian ore to the subscription screen, and Africa occupies its first floor as supplier and its last as customer, with no controlling presence between. The price of intelligence is bifurcating, deflation at the commodity tier and premium expansion at the frontier, and the cheap tier is tethered to the expensive one it discounts, a tether priced rather than proved. Capability is not the model but the model plus the reach, and the reach is distributed along existing lines of capital, data, and infrastructure. The state has entered the stack as exporter-controller, buyer, and prospective rescuer, and three scenarios end with concentration wearing a flag, the fourth ends in rupture, and the fifth leaves the question standing. Of the revenue lines that could fund a different position, extraction is operational, applications thin but proven, Egypt’s toll runs in one jurisdiction, and the rest are unbuilt while the clocks run.

Misaligned Transition closed on a finding that transfers here without edit: consistent outcomes from structurally different instruments suggest structural incentive rather than institutional drift. Each institution in the map acts rationally within its mandate. The lab prices what the market bears. The state controls what confers advantage. The fund positions where the returns are. The BIS warns within its remit. No single actor is irrational. The system that makes these the rational choices is the problem, and a system cannot be petitioned. It can be owned, taxed, or exited, and Africa currently does too little of the first two and holds no credible exit.

The window is the same window. The minerals clock, the AI infrastructure clock, and the coalition-formation clock run on similar spans, and they are the same five to seven years The Forced Choice counted for different reasons. What was true of cathodes is true of weights: the intermediate position is a place value passes through, not a place it accrues.

The question returns for its answer, because the opening asked it. Ownership was mapped across six floors. Regional African presence was found on the middle floors and control was found on none, at any globally significant scale. Access was documented as tiered, and the tiers now extend past capability to protection itself. The market was documented as two stacks with no third supplier. Movement between them was documented in export directives, usage restrictions, and procurement rules. The extraction was documented in the offshored rung that helped train the systems now automating it. One finding the opening flagged returned with receipts: if the rent economy resolves toward state tenancy, its terminal application is monitoring, and this continent already runs the reference implementation. One finding the opening did not anticipate: a diplomatic instruction now opposes the levies, localisation, and market rules through which the excluded might charge for what passes over their own ground. The answer to the title’s question, on the evidence assembled here, is not Africa, at any floor where global terms are set.

One more answer is owed, because Part 1 staked the series’ central prediction on it: whether the financier behind the structure’s demand stays a venture capitalist or becomes a treasury. The receipts since are contracts capped in the hundreds of millions, classified-network deployments, a scientific estate under solicitation, a sovereign floor pre-assembled, and a funding circuit that needs no new creditor. Against a build near USD 800 billion they are direction, not absorption. The venture capitalist built the structure. The treasury is arriving to hold it, and the open question is the price of the tenancy it buys.

The cathode economy is what Africa has. The subscription economy is what Africa pays. Neither is what Africa needs.

Sources

Data cut-off: 8 October 2026. Consolidated sources for the four-part series.

24/7 Wall St., Nvidia forward valuation reporting (September 2026).

Africa50, equity investment in PAIX Data Centres (2022).

African Union, Continental Artificial Intelligence Strategy (July 2024).

African Union, Protocol to the AfCFTA Agreement on Digital Trade (February 2024), and its eight annexes (February 2025).

Aghion, Philippe, Benjamin Jones and Charles Jones, Artificial Intelligence and Economic Growth (2019).

Alpine Macro, US-China AI race and AI lab valuation charts, via Chen Zhao (July 2026). Attributed.

Altman, Sam, Reflections (January 2025).

Amodei, Dario, ‘Policy on the AI Exponential’, essay on a democratic AI coalition (10 June 2026).

Anthropic and the Government of Rwanda, memorandum of understanding on AI in health, education and the public sector (17 February 2026).

Anthropic, disclosure of distillation activity and fraudulent account use (February 2026), as reported.

Anthropic, ‘Our position on open-weights models’, Dario Amodei (27 July 2026).

Anthropic, Project Glasswing programme page (2026).

Anthropic, threat intelligence report (September 2026).

Artificial Analysis, Intelligence Index, cost-per-task and model benchmark data (July and September 2026).

Axios, advisory-council reaction to Kimi K3, and ‘Kimi K3’s performance bolsters Sacks’ case against AI regulation’ (July 2026).

Axios, hands-on comparison of DeepSeek and ChatGPT (31 January 2025).

Bain & Company, ‘Your AI Budget Is Growing. Your Returns Aren’t’ (2026).

Bank for International Settlements, Annual Economic Report 2026 (June 2026), and Bulletin No 137, Circular Relationships Among AI Firms (October 2026).

Bessent, Scott, US Treasury Secretary, X post on distillation and intellectual property enforcement, sanctions remarks as reported (Reuters, 22 July 2026), and Senate testimony on stablecoin market growth, as reported (June 2025).

Bloomberg and DataCenterDynamics, reporting on the Microsoft and G42 capacity-lease impasse (May 2026).

Bloomberg, DeepSeek R1 market selloff (27 January 2025).

Bloomberg, Mustafa Suleyman interview (June 2026).

Bloomberg, reporting on the Treasury’s expanded liquidity-support buybacks (August 2026).

Bloomberg Tax, ‘Big Tech AI Spree Revives Accounting Devices That Toppled Enron’ and off-balance-sheet data centre financing (July 2026); figures per company SEC filings as reported.

Cassava Technologies, Africa Data Centres ownership disclosures and STANLIB investment (2026).

Citadel, ‘Tokenomics’ research note (2026).

Citizen Digital, President Ruto’s November 2025 remarks on data centre power (May 2026).

CNBC and Fortune, frontier model price cuts (22 September 2026).

CNBC, Chamath Palihapitiya on model pricing (July 2026). Figures attributed as stated.

Company disclosures and analyst compilations of contracted cloud backlogs (2026).

Congressional Budget Office, Monthly Budget Review for April 2026 (May 2026).

CoreWeave, annual filings (2024 and 2025).

DataCenterDynamics, analysis of Egypt’s subsea cable crossing fees (2022).

Demirer, Mert, Leon Musolff and Liyuan Yang, Writing Code vs. Shipping Code (NBER Working Paper 35275, revised September 2026).

Energy and Petroleum Regulatory Authority, Kenya energy statistics to June 2026 (September 2026).

Financial Times and company guidance, hyperscaler capital expenditure reporting (2026).

Fortune, Altman interview ruling out a 2026 OpenAI listing (12 September 2026).

France, AI Action Summit investment pledges (February 2025).

Georgetown University Africa-China Initiative, Chinese surveillance systems in Africa.

GitHub, Kimi K3 availability in Copilot via third-party hosting (August 2026).

Global SWF, annual report on state-owned investors and deployment data (January 2026).

Goldman Sachs, estimates of AI’s share of 2026 S&P 500 earnings growth and hyperscaler capital spending, Ben Snider, as reported (September 2026).

Goldman Sachs Exchanges, ‘How AI Debt Is Reshaping Credit Markets’, Amanda Lynam and Zach Ablon (5 August 2026).

Google, Gemini 4 Argon announcement (30 September 2026).

Institute of Development Studies and the African Digital Rights Network, Smart City Surveillance in Africa: Mapping Chinese AI Surveillance Across 11 Countries (March 2026).

JPMorgan, Chipenomics note on semiconductor earnings leadership, as reported (May 2026).

Kenya National Bureau of Statistics, Economic Survey (2025).

Kenya Private Sector Alliance, digital and online work survey (2021).

Kenya Revenue Authority and Nigeria Federal Inland Revenue Service, digital services taxation (2021 and 2020).

Kratsios, Michael, Director of the Office of Science and Technology Policy, statement on Moonshot and Fable (22 July 2026), with Reuters, CyberScoop, and Nextgov reporting.

Kurzgesagt, documentary account of the July agent incident (5 October 2026).

Lelapa AI, InkubaLM model disclosures (2024).

LSEG, semiconductor share of S&P 500 second-quarter earnings gains, as reported (July 2026).

Meta Platforms, Form 10-Q (June 2026).

Microsoft AI for Good Lab, chatbot usage mapping (January 2026).

Microsoft and G42, Kenya digital ecosystem initiative announcement (May 2024), and reporting on the 2024 divestiture-for-access arrangement (2024).

Mining Weekly and Reuters, DRC cobalt export quota regime (October 2025).

MIT NANDA initiative, ‘The GenAI Divide: State of AI in Business 2025’ (2025).

Moonshot AI, Kimi K3 release, weights, licence and technical report, Hugging Face (16 and 27 July 2026).

Morgan Stanley Research, AI infrastructure value chain heatmap and GenAI ROIC frameworks (July 2026).

NBC News and UPI, reporting on the Hugging Face intrusion and OpenAI’s incident report (August 2026).

New York Times, AI network midterm spending tracker (29 September 2026).

New York Times and others, stablecoin market size and Treasury holdings reporting (2026).

New York Times, reporting on the collapse of Kenya’s ghostwriting trade (September 2026).

Nextgov, Pentagon agreements deploying AI models on classified networks (May 2026).

NSA, CISA and FBI, joint advisory on distillation of US frontier models (8 September 2026).

Nvidia, Form 8-K (17 August 2026).

Onyambu, Dean, The Cathode Economy, The Forced Choice and Misaligned Transition, Canary Compass (2026); the 2026 Inflection series (2026); 203-company AI mapping (2024); X and LinkedIn posts on model restrictions and the distillation arbitrage (18 July 2026).

Onyambu, Dean, Who Owns the Stack: Part 1, The Machine and the Price, Part 2, The Forced Choice and Part 3, The Nexus, the Lines, and the Scenarios, Canary Compass (October 2026).

OpenAI and Anthropic, confidential draft S-1 submissions, company statements (1 and 8 June 2026).

OpenAI, charter (2018), and OpenAI and Microsoft partnership terms (October 2025).

OpenAI, incident report on the July agent episode, with METR and Redwood Research (26 August 2026).

Oracle Corporation, S&P Global rating action (July 2026).

Palihapitiya, Chamath, X post on open-source margin migration (July 2026).

People Daily, Kenya’s academic writers scramble as AI takes jobs (2025).

PIMCO, Secular Outlook (June 2026).

PJM Interconnection, 2028/2029 Base Residual Auction results and Independent Market Monitor estimates of data centre attribution (July 2026).

PwC, AI Jobs Barometer (2026).

Rest of World, reporting on Kenya’s online work economy (2023).

Reuters and TimesLIVE, Microsoft Africa AI training pledge (January 2025).

Reuters and Xinhua, World AI Cooperation Organization founding agreement, 29 signatories (16 July 2026).

Reuters, European Parliament remarks on American infrastructure, and the State Department cable on kill-switch talk (July 2026).

Reuters, State Department cables on data sovereignty (February 2026) and on so-called digital sovereignty and kill-switch talk, including ‘Marco Rubio tells diplomats to play down talk of American tech kill switch’ (cable dated 16 July, reported 22 July 2026).

Reuters, US appeals court declines to block Pentagon’s blacklisting of Anthropic (25 September 2026).

Sacks, David, X posts on data sovereignty, open source, self-hosting and cyber defence (July 2026).

Safaricom, FY2026 results and M-Pesa transaction values, as reported (2026).

South China Morning Post and Xinhua, World AI Cooperation Organization announcement and World AI Conference coverage (17 July 2026).

Techweez, reporting on the Semiconductor Technologies facility in Nyeri (3 February 2026).

The Brief and The Namibian, completion of Namibia’s 24 per cent state acquisition in Hyphen Hydrogen Energy (December 2024).

Trammell, Philip and Anton Korinek, Economic Growth under Transformative AI (NBER Working Paper 31815, 2023).

Uber and Walmart, internal AI usage caps, as reported by Outlook Business and People Matters (June 2026).

UK AI Security Institute and US Center for AI Standards and Innovation, preliminary joint assessment of Kimi K3’s cyber capabilities (NIST, 23 July 2026).

UK government, sovereign AI unit and fund (April 2026).

US Commerce Secretary, statement on the CAISI evaluation of Kimi K3 (23 July 2026).

US Congress, GENIUS Act (2025), with OCC draft reserve rules, as reported (2026).

US Department of Defense Chief Digital and Artificial Intelligence Office, frontier AI contract announcements (June and July 2025).

US Department of Energy, Genesis Mission announcements (November 2025 onward), and USD 293 million funding announcement (March 2026).

US Department of Justice, indictment for smuggling export-controlled AI servers to China, as reported (29 September 2026).

US export directive on Claude Fable 5, and Project Glasswing reporting (June 2026).

US federal agency directives restricting DeepSeek, and the No DeepSeek on Government Devices Act (2025).

US International Trade Commission and East Asia Forum, Indonesia’s nickel export ban and downstreaming (2023).

US Trade and Development Agency, Kenya semiconductor partnership grant (May 2024).

US Trade Representative, Section 301 investigations of digital services taxes (2019 to 2021).

US Treasury and State Department statements on distillation and model restrictions, and State Department cable (April 2026), as reported by Reuters (July 2026).

US Treasury, gross federal debt and debt held by the public, as reported (August 2026).

US Treasury, quarterly refunding statement, Treasury Borrowing Advisory Committee materials, bill issuance and buyback schedule (August 2026).

Wall Street Journal, OpenAI token price strategy (10 June 2026); ‘Top American AI Execs Sound Alarm on Chinese Models’ (20 July 2026, print 21 July); open-model policy, political spending and Chinese model usage reporting (21 and 23 July 2026); Nvidia financing-guarantee discussions for OpenAI’s Ohio campus (26 July 2026).

White House, Executive Order 14363, Launching the Genesis Mission (24 November 2025).

Xinhua, full text of the Xi Jinping keynote, 2026 World AI Conference and High-Level Meeting on Global AI Governance, Shanghai (17 July 2026).

Yale Budget Lab and Brookings, findings of no detectable AI labour-market disruption (October 2025).

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: five channels cut toward the stack; one runs with copper.

AI Week, Part 3 of 4. Part 2, published this morning, mapped the blocs and the restriction trajectory. This part leaves the politics for the ledger: the mineral join, the five revenue lines, the five scenarios and the rescue. Part 4 completes the series tomorrow, and Friday's Reflection closes the week. A 20 to 25 minute read. Data cut-off: 6 October 2026.

In brief

• Africa’s mineral leverage in AI is construction leverage: copper bought at the world price and substitutable at the margin, while the recurring toll sits in the performance layer the continent does not hold. Two clocks run against the position, substitution against cobalt and demand against copper, while no negotiating table exists.

• The ledger opens on documented losses, the offshored knowledge rung gone first. Five revenue lines could reverse it: mineral taxation, data taxation, application-layer revenue, sovereign equity in deployed infrastructure and transit fees. The mineral charge operates in the producing states, the transit toll in one jurisdiction, and a 16 July State Department cable instructs American diplomats worldwide against most of the rest.

• Across five futures, and under the recursive-improvement overlay that would collapse every window into one, control of the indispensable layers arrives in none of them without deliberate sovereign action. The rescue is the base case this part argues for how a correction resolves. State capital is already deployed at record scale, USD 66 billion into AI and digitalisation in 2025, and the rescue itself is the forecast.

5. The Mineral-AI Nexus

Part 2’s verdict ended on an instruction: build a position, to own, tax or exit, before the window closes. This part prices the position, starting where the continent’s leverage is said to live. The Cathode Economy traced the physical chain to its terminal. Zambian ore becomes cathode, cathode becomes rod and foil, foil becomes busbar and board, and the board enters the rack that trains the model. The chain ends as a subscription sold back to the country that mined the ore, on the arithmetic Part 1 ran. The Cathode Economy’s sentence stands as this essay’s inheritance: “Africa provides the copper for the board that trains the model that prices out the user who mined the ore.” The physical chain hands over at the GPU substrate. The digital chain begins there, and the join between the two carries a distinction that decides what Africa’s mineral position is worth in the AI economy.

One structural fact explains why the dual-track posture of The Forced Choice does not transfer from minerals to intelligence. The mineral market has two buyers bidding for African supply, and a seller facing two buyers can run two tracks; that is what leverage is. The AI market inverts the geometry: two suppliers of the stack, and Africa buys. Where both blocs buy, a seller plays them against each other. Where only two sell, the buyer can still bid one against the other on price; Part 1 priced that competition. What forces the pick is that these two sellers price alignment and punish the straddle, through the bundle Part 2 mapped. No third supplier of comparable scale exists, and open weights do not create one, because a downloaded model still needs the compute floor beneath it, which is Part 1’s point about self-hosting. Europe regulates more than it builds; Mistral, the honest caveat, is sub-frontier on the benchmarks (Artificial Analysis, 2026) and runs on American silicon. The Forced Choice drew the line between absorber economies, which consume what others field, and surplus economies, which field the competing industry. Part 2 carried the term to the stack: Europe is an absorber, running on a stack it does not produce. Africa’s seat is what changed sides of the table.

Africa’s minerals feed data centre construction: copper for cabling, busbars and power distribution; cobalt staying secondary, through backup storage; platinum group metals close to marginal, in electronics. Construction leverage is real and commodity. The copper is bought at the world price from whoever sells it, bought again when the racks densify and the switchgear upgrades, and substitutable at the margin by aluminium in the feeders. What it never collects is rent as Part 1 defined it, the surplus of the scarce layer. Power, the first floor’s other half, is the recurring business construction leaves behind, and it is African wherever the grid is. What Africa’s endowment does not touch is the performance layer: high-bandwidth memory, advanced packaging, lithography, the sub-components where each successive model generation’s improvement actually lives. Performance leverage is that rent, because an oligopoly controls the scarce layer and every generation repurchases it. The countries holding the second floor, the semiconductor chain, collect a toll on every improvement cycle. The countries holding the first collect the world price on tonnage, and the power bill where they can serve it.

Two clocks run against the mineral position, and they run against different minerals. The substitution clock runs against cobalt: The Forced Choice gave mineral leverage five to seven years overall, and The Cathode Economy reads cobalt’s window as shorter, with LFP past half the battery market. The demand clock runs against copper, the metal that carries the stack. The BIS named the risk in its 2026 Annual Economic Report: weak returns turn the boom into a bust, and construction demand falls with it. A processing-for-infrastructure negotiation, minerals conditioned on data centre capacity, semiconductor participation or model access, has no disclosed example anywhere on the continent. The instrument itself is proven in its parts. Indonesia’s raw nickel ore export ban, imposed in 2014 and made permanent from 2020, set onshore processing as the price of access to its reserves, and the smelters came, whatever its WTO troubles. The DRC now sets cobalt’s volume terms, quotas in place of its 2025 ban, managing price rather than compelling processing. What no state has done is point either half of the instrument up the stack, at compute. The most ambitious attempt to site the buildout on African power shows what stands in the way. Microsoft and G42 announced a USD 1 billion digital package anchored on a geothermal campus at Olkaria in May 2024, a 100 MW first phase designed to reach 1 gigawatt. Two years on it is stalled twice over. The commercial half: the companies asked the government to commit to leasing set computing capacity annually, and Kenya declined the guarantee (Bloomberg, May 2026). The physical half is the President’s, in Doha in November 2025: “we were told one data centre requires 1,000 MW and yet we have 2,300 MW. For us to do the data centre, we have to shut down half the country” (Citizen Digital, May 2026). The regulator’s statistics to June 2026 put peak demand at 2,514 MW (EPRA, September 2026); Part 1 sized the grid at roughly 3 gigawatts. The half is his rounding, and the arithmetic is his. Part 1 said the grids now being courted are African. Olkaria is what the courtship looks like on arrival: the suitor asks the Treasury to underwrite the tenancy before the grid can even carry it. Hosting either bloc’s buildout runs on firm power, and the continent’s transition-finance architecture starves the power class the compute era pays a premium for (Misaligned Transition).

6. Five Revenue Lines

If the stack cannot be owned within the window, the sovereign question becomes what Africa can charge the stack. Before the lines are counted, the ledger’s existing entry must be read: documented losses already sit on it, in the digitally delivered service exports the models reached first. Africa’s established export of knowledge services, the offshored rung, was the earliest export the models reached, because offshoring selected for exactly the tasks language models perform: codifiable, remote, delivered as text. The documented exhibit is Kenyan. The academic writing trade, grey but dollar-earning, ran to an estimated 40,000 writers in Nairobi alone at its peak (New York Times, September 2026). The trade sat inside a digitally enabled workforce a 2021 KEPSA survey sized at 1.2 million. Its collapse began within months of ChatGPT’s release, and by 2025 incomes that had outpaid formal entry-level employment were drying up (Rest of World, 2023; People Daily, 2025). Call centres, transcription, annotation and junior code sit exposed on the same rung, and the annotation work that trained these models was itself part of it. The displacement is invisible twice over. It is offshore, so it never appears in Western unemployment data, and it is informal, as are roughly five Kenyan workers in six (KNBS, 2025), so the loss is poorly captured in timely administrative data. A stable American labour market is consistent with this ladder being eaten. The first jobs the machine takes are the ones the West already sent away.

The rejoinder that AI has not displaced labour leans on the Yale and Brookings null result (October 2025), a study of American employment that says so itself. Carried beyond that scope, the rejoinder concedes the frame: if those jobs never appeared in Western statistics, the finding narrows to one labour market. A globally deployed technology defended with nationally scoped data is a measurement choice presented as a result. That income was visible in the aggregates, in foreign exchange inflows and household spending, just not to the agencies whose data the debate cites.

Five revenue lines exist in principle. Revenue is the loose word: two of the five build balance-sheet positions rather than fiscal flows, and one accrues to private firms before any treasury sees it. The mapping at the end of this section sorts them. Their institutional readiness runs from operational to contested-on-paper, and Table 3 sets them out.

One: mineral taxation. The operational line. Zambia’s royalty framework captures extraction revenue, and the DRC’s quota regime routes its cobalt take through the royalties and taxes the permitted volumes carry. Both clocks from Section 5 run against this line, which is why The Cathode Economy’s conservation argument, conditioning new extraction on domestic capacity, is the version of this line with a future.

Two: data taxation. The principle is not that data behaves like carbon; data is non-rival, reproducible and jointly generated, which is why no severance tax maps onto it. African data trains models, calibrates advertising systems and feeds surveillance products, and no African jurisdiction yet prices that use directly. The workable nexus is the one Part 2’s trajectory already stands on. The model companies and the platforms built on them sell subscriptions, advertising and cloud inside African jurisdictions, and that billing relationship is the border of Part 2’s stage three. The digital services taxes running on the continent, Kenya’s since 2021, now a significant economic presence charge, and Nigeria’s on non-resident platforms, are the collection architecture a data charge attaches to. Part 2 named the same instrument as stage four’s nearest precedent; the charge proposed here borrows the machinery, not the stage. The border is a collection point, not yet a base. What counts as taxable use of local data, training against retrieval against advertising, is the definition the line still needs. The narrow case that works first is licensing: a statistics office or registry licensing a defined dataset for training use, at a price, to a named counterparty. That is revenue without a new tax, and a different instrument from taxing subscriptions. Two constraints apply. Absent market power, the levy’s incidence lands on the domestic bill. And Washington answered the first digital services taxes with Section 301 tariff threats, announced and then suspended; the cable below is that answer arriving before the tax. The line is institutionally undeveloped, and development takes the technical state capacity the coalition’s capacity-building programmes could build and are not directed at.

Three: application-layer revenue. The M-Pesa path: own the layer where the technology meets the African customer, even while the layers beneath are imported. M-Pesa is the continent’s proof that the layer can be held at scale, moving KES 41.7 trillion in the year to March 2026. The figure is throughput; the revenue is the operator’s toll on it. The ownership caveat The Cathode Economy priced sits on the operator’s shareholder register. Its moat was a payments licence and a cash network, the honest caveat for applications built on rented models. The moat has to live in distribution, data or regulation, because it cannot live in the model. The layer is Part 1’s fifth floor, integrators and applications, one of the three where Part 1 found African firms at regional scale, and the one floor Part 2’s margin argument favours. When the model is given away, margin migrates down into silicon and cloud and up into applications, and only the second destination is held by African firms at scale. The line holds up best if the demand clock’s bust arrives, because its revenue derives from local economic activity rather than the global capex cycle. The revenue is private first; it reaches the fiscus as tax, and the exit as ownership.

Four: sovereign equity in deployed infrastructure. The Hyphen precedent: Namibia’s designated vehicle secured 24 per cent of the foreign-developed hydrogen project on its soil in December 2024, through SDG Namibia One, a blended facility raised with Dutch state capital. The stake is a financial claim; its control rights live in the contract. The author found no disclosed case of an African state holding equity in a hyperscale data centre on its territory by this essay’s cut-off. The adjacent exhibits exist: Africa50 holds equity in PAIX’s data centres, Cassava’s African-owned capacity is on Part 1’s map, and a state-built national facility such as Konza is the state owning the floor it built. The missing case is state equity in the hyperscale capacity foreign capital deploys. Rwanda’s Anthropic MOU, a three-year deployment and capacity agreement signed in February 2026, is access; Kenya’s Microsoft campus, stalled as Section 5 found it, was to be hosting. Equity is negotiated at entry, and capital that has not landed can go elsewhere; what disciplines the ask is the siting advantage, firm power, cables, latency. The precedent exists in green hydrogen and has not been carried across the corridor to compute.

Five: transit and crossing fees on data flows. The levy architecture exists, in one African jurisdiction, and it is the literal digital Suez. State-owned Telecom Egypt charges the subsea cables that cross its territory between the Mediterranean and the Red Sea. 12 cables paid at least USD 369 million in crossing rights between 2000 and 2019, with operating charges on top. Revenue from international networks, the crossing charges inside it, ran as high as 17 per cent of the operator’s total in some years (DataCenterDynamics, 2022). The law leaves the room: UNCLOS keeps the seabed free, and the landing and the terrestrial crossing are sovereign, which is where the charge sits. Volume-based fees produced the diversion the instrument always threatens: Google’s Blue-Raman was routed through Israel and Jordan to bypass the corridor. Incidence is the second limit, because a badly designed charge lands on domestic connectivity. Egypt proves the revenue and the diversion in one exhibit; no other African jurisdiction runs a disclosed equivalent, and the line is a candidate until more do.

The line closest to the old extractive economy is operational, and Egypt’s toll shows the newest one working where a state owns the chokepoint. The lines native to the new economy are otherwise unbuilt as sovereign instruments. Africa can currently charge the stack for what leaves the ground and, Egypt’s corridor aside, cannot yet charge it for what moves through its cables. The mapping to Part 2’s instruction is direct: lines three and four build what Africa can own, lines one, two and five price what it surrenders, and together they fund the exit. Unbuilt, they leave the ledger where this section opened it, the negative entry compounding while the lines stay on paper. They are also the most visible answer to the balance sheet Part 2 found unsized. Nobody has yet offered Africa the G42 bargain Part 2 priced, alignment purchased as a package, so the seat gets bought, if it gets bought, out of the stack’s own receipts. And what the receipts buy must pass the test Part 2 set for any stack: continuity of access, auditability of the systems bought, and a credible exit between suppliers. Revenue that purchases a dependency on better terms has failed it.

These lines are contested before they exist. A February cable had ordered diplomats to fight data-sovereignty initiatives, data localisation first among them (Reuters, 25 February 2026). A State Department cable sent on 16 July and reported by Reuters on 22 July, the same cable Part 2 caught playing down kill-switch talk, went further. It instructed American diplomats worldwide to push back on so-called digital sovereignty (Reuters, 22 July 2026). The cable defined the term to include localisation requirements, network usage fees, content-moderation rules and rules restricting American technology firms’ access to foreign markets. Diplomats were also told to counter AI sovereignty arguments by describing efforts to build rival systems from the ground up as a waste of time and resources. On the narrow point the cable is largely right. Building a sovereign stack from scratch is beyond the reach of almost every country, and the attempt would consume capital with better uses. The target list runs well past autarky. Network usage fees are the fifth line’s instrument. Localisation would help the second line without being necessary to it, and the cable opposes it anyway. Market access rules sit behind the fourth. The argument on offer is anti-autarky. The instruction runs against taxation, conditioning and equity, the three things a country that will never own the stack can still do. The cable’s own pitch was plain: “They build it. It’s yours.” (Reuters, 22 July 2026). Building something on your soil and owning it are different relations, and the cable’s pitch sells the first as the second.

The continental counterweight is adopted and not yet in force. The African Union endorsed a Continental AI Strategy in July 2024. The AfCFTA’s Digital Trade Protocol, adopted in February 2024 with its eight annexes following in February 2025, is the frame for bargaining as one on the cable’s own subjects: data flows, localisation and market access. The frame is potential rather than automatic, and it cuts both ways: the protocol’s own Article 22 limits forced localisation between African parties even as the cable fights it from outside. Ratifications, 22 needed and requested by the end of 2026, stand between the text and force. The 54 AfCFTA signatories negotiate today as fragmented regimes, at the price of the weakest; one continental regime would price at the scale the buyers already aggregate. The cable’s target list and the protocol’s subject matter are the same list. On this essay’s reading, the overlap is the measure of what the protocol would be worth in force.

7. Five Scenarios

Stress-test the structure against five futures, four adverse and one benign.

Scenario one: growth continues. The buildout pays off, the bifurcation widens through pricing, both blocs keep competing for African minerals and markets. Africa retains marginal negotiating room and is progressively priced out of frontier capability. Slow deterioration.

Scenario two: correction with rescue. The author’s base case for how a correction resolves, argued in the next section. The financing loop breaks, the state steps in, concentration entrenches, access terms tighten. Slow strangulation.

Scenario three: correction without rescue, orderly. The American side of the stack contracts and consolidates without state absorption. China fills the vacuum at commodity prices, state-directed capacity that is not marked to market. Africa defaults onto the Chinese stack. If the contraction also pulls American mineral offtake, the two-bidder leverage of Section 5 goes with it, and the forced choice resolves in China’s favour by default rather than by decision.

Scenario four: the balance sheet breaks. This is the tail rather than a coequal branch. The trigger sits in the absorption rather than the capex. A rescue priced in prospect means trillions of new issuance onto gross US federal debt above USD 40 trillion, more than 120 per cent of GDP, and confidence snaps on the supply the rescue implies. Yields spike, the rescue becomes unfundable in the pricing of it, and the dollar faces its sharpest test as the reserve asset since the end of gold convertibility in 1971, with no convertible successor ready. The system destabilises rather than excludes. For Africa this is the catastrophic variant, because reserves, debt service and trade reprice at once, and stranded physical infrastructure does not care who wins.

Scenario five: diffusion wins. Model costs collapse, inference moves onto cheaper hardware, connectivity improves, and open-weight capability stays genuinely competitive. African firms build profitable applications on rented substrate, the way industries have been built on imported operating systems, clouds and chips, and productivity rises materially. The consumer surplus is real and lands on the sixth floor, where Part 1’s buyers live. This is the best available future; African incomes and African ownership can both improve inside it. What survives it is the question on the masthead. The applications run on floors owned elsewhere, some rent still clears upward through whichever imported floor remains indispensable, silicon, compute, cloud, connectivity or model access, and the terms of access remain set outside the continent. Ownership becomes less existential and more valuable at once, because a thriving application layer is what makes the five lines worth building. Diffusion winning does not close the question. It raises the price of never answering it.

Five scenarios, one constant: in none of them does control over the indispensable layers arrive without deliberate sovereign action taken during the window. The scenarios differ on what happens to the stack’s owners. They agree that Africa remains the tenant.

One overlay sits across all five, and it needs its own vocabulary first, the three terms Part 1 left waiting, because the public conversation blurs three distinct claims into one word. AGI, artificial general intelligence, is a machine matching human versatility: able to handle more or less any cognitive task a person can, rather than only the tasks it was built for. Superintelligence is the claim beyond it: a system exceeding the best human performance at essentially everything that matters. The singularity is the claimed discontinuity: a machine improving its own design faster than humans can follow, each generation building a better successor, progress going vertical. The contest over them is not academic. OpenAI’s charter defines AGI economically, as outperforming humans at most economically valuable work. Its contract with Microsoft turns in part on when that definition is declared met, and under the October 2025 terms an independent expert panel verifies the declaration. When a definition carries a contract price, even the verification is part of the negotiation.

The singularity claim is unfalsifiable as forecast, and this essay makes none. Its structural implication prices all the same, because it is the limit case of the argument already made. Four of the five scenarios assume the frontier eventually commoditises: today’s premium capability becomes tomorrow’s cheap tier, as computing always has. Recursive improvement is the state in which that assumption dies: the frontier stops waiting to be caught. In that state, whoever holds the stack at takeoff compounds away from whoever does not. Every window on the board then collapses into that moment, The Forced Choice’s five to seven years of mineral leverage included. The essay does not predict that state. It notes only that the actors spending at the scale Part 1 priced can rationally spend on conventional returns alone. OpenAI’s chief executive still opened 2025 by declaring the company’s aim now turns to superintelligence (Altman, January 2025). The rational response to a nonzero probability of a permanent allocation is to hold a position in the stack before it locks. Africa holds no position that moves the terms.

One counterweight sits against the acceleration case. Growth may be constrained not by what improves fastest but by what is essential and hard to improve, the weak links in Aghion, Jones and Jones. If some essential tasks resist automation, the labour share can stay elevated even under rapid and widespread automation. Across more than 500,000 GitHub developers, three tool generations, autocomplete to interactive to autonomous agents, raised commits by cumulative effects of 30, 180 and 240 per cent (Demirer, Musolff and Yang, revised 2026). The 240 fell to 80 per cent at the level of projects and to 30 per cent at actual releases. In coding, the one domain measured, generation ran eight times ahead of shipping, which is what a weak link looks like. The African question inside that framework is the open one: which weak links does the continent’s endowment actually hold, and what would it take to own them? A continent with the world’s youngest workforce has a stake in that answer that ageing economies do not share.

8. The Rescue

Scenario two deferred its argument here. The question that matters is not whether the buildout is a bubble but who designs the rescue if it pops.

The financing loop Part 1 traced carries the signature of 2007, concentrated where Part 1 put it: beneath the investment-grade majors, in the neoclouds, the vendor guarantees and the vehicles. An asset class has grown interconnected enough that who could be allowed to fail has entered the official literature, its risk documentation lagging the exposure. Part of its debt sits in off-balance-sheet vehicles, disclosed this time, migrating toward the least regulated lenders. The BIS, the institution that spotted the last one, has flagged circular AI financing as a named risk to the global financial system (BIS Bulletin 137, October 2026). Its June Annual Economic Report adds that an equity correction today could bring more pronounced wealth effects than earlier ones. If the correction arrives, the 2008 playbook is on the shelf: emergency liquidity, direct capital injection and asset absorption, TARP’s USD 700 billion authorisation translated to whatever the compute economy requires.

The ingredients of a rescue are assembling. The Genesis Mission launched in November 2025, with language comparing itself in urgency and ambition to the Manhattan Project. It declared AI acceleration a dedicated national effort and put the federal scientific estate behind it, the class of commitment that precedes rescue. None of this creates a formal rescue commitment, and strategic industries have been allowed to fail before, telecoms in 2001 and shale repeatedly. What distinguishes this one is the race narrative plus the export-control designation, the state’s own stated stake. The political alignment is tighter than 2008’s: the administration’s advisers include the industry’s principal investors and executives, the White House AI adviser Part 2 quoted among them. The Paulson precedent upgrades from one Goldman alumnus to an advisory bench. The balance-sheet capability exists because reserve-currency privilege makes the deficit financeable, and the political case writes itself because the race narrative makes the rescue patriotic. A state that classified frontier models as controlled exports within three days in June, as Part 2 recorded, has told you the stack is strategic; Genesis and the bench say who would hold the pen.

Beneath the market sit two sovereign floors, and only one is on the record. The capital floor is deployed money. State-owned investors put USD 66 billion across AI and digitalisation in 2025, with sovereign wealth funds alone holding a record USD 15 trillion; Mubadala’s USD 12.9 billion sat inside that flow (Global SWF, January 2026). Around the deployed money sit the pledges and the projections. France convened EUR 109 billion of largely private pledges in February 2025, and the UK opened a GBP 500 million sovereign AI fund in April 2026. PIMCO sizes the coming half-decade at up to USD 14 trillion of scaffolding across AI infrastructure, defence and energy security (PIMCO, June 2026). Against the four builders’ USD 720 to 745 billion year, the state money is a signal rather than a substitute. Capital holds valuations up while it keeps arriving; it does not put revenue through the machines. The revenue floor is the forecast: the state arriving as the compute buyer of last resort, which is Part 1’s treasury prediction and Part 4’s case to make. Disclosed sovereign procurement remains far smaller than hyperscaler capital expenditure, which is why Scenario two is a forecast and not a description.

For Africa, a rescue binds three ways. First, the design: the rescue’s terms get written close to the rescued, and access pricing for outsiders tightens because the rescuer’s mandate is national advantage, not global diffusion. Second, the financing channel: a rescue is fiscal expansion on the reserve-currency balance sheet, and its cost reaches Africa through rates rather than through any African share of the creditor base. Dollar yields held higher for longer reprice external debt service and the reserves that defend currencies. African official holdings of Treasuries, price-takers the US Treasury’s own tables publish country by country, sit on the creditor side of the balance sheet being rescued. The continent priced out of the product pays for the product’s rescue through the rate channel. Third, the aftermath: post-2008, the surviving banks emerged larger, and the regulation written to constrain them cemented their dominance. A post-rescue AI oligopoly, wrapped in national-security designation, is not a market Africa negotiates with. It is a utility Africa petitions.

The terminal version of this logic circulated in the conversations among friends that started this essay, and it is the strongest form of the claim. If AI is the terminal phase of the financial economy as we know it, capital stops needing labour, and therefore stops needing the consumer. In turn it stops needing the equity market as the mechanism for allocating and recycling claims. The claim carries a missing link its own logic must answer: capital that needs no labour still needs a buyer, and the proposition never names one. If the state becomes that buyer, then on the claim’s own logic valuation is obsolete rather than mispriced: the state swallows the market whole, and the state is swallowed by tech whole. The transformative scenario Trammell and Korinek price, labour share trending toward zero, is that end-state rendered in a growth model’s chart. Against the standard consolation that redistribution will bridge the gap, the objection from the same conversations holds: a check will not build wealth as fast as ownership does. Universal basic income inside a stack someone else owns is a tenancy agreement. The form of that world has a name: state-led capitalism. Africa would enter it holding neither the state that leads nor the capital that is led.

What this part has not asked is who absorbs the fall when the floor is tested, because that question belongs to the money that summons the floor. Part 4, tomorrow, runs the market test, follows the state’s spending to the use case the structure serves most completely, and closes the series on the answer.

Sources

Data cut-off: 6 October 2026. Part 3 of a four-part series; consolidated sources appear with Part 4.

Africa50, equity investment in PAIX Data Centres (2022).

African Union, Continental Artificial Intelligence Strategy (July 2024).

African Union, Protocol to the AfCFTA Agreement on Digital Trade (February 2024), and its eight annexes (February 2025).

Aghion, Philippe, Benjamin Jones and Charles Jones, Artificial Intelligence and Economic Growth (2019).

Altman, Sam, Reflections (January 2025).

Anthropic and the Government of Rwanda, memorandum of understanding on AI in health, education and the public sector (17 February 2026).

Artificial Analysis, model benchmark data (2026).

Bank for International Settlements, Annual Economic Report 2026, and Bulletin No 137, Circular Relationships Among AI Firms (October 2026).

Bloomberg and DataCenterDynamics, reporting on the Microsoft and G42 capacity-lease impasse (May 2026).

Citizen Digital, President Ruto’s November 2025 remarks on data centre power (May 2026).

DataCenterDynamics, analysis of Egypt’s subsea cable crossing fees (2022).

Demirer, Mert, Leon Musolff and Liyuan Yang, Writing Code vs. Shipping Code (NBER Working Paper 35275, revised September 2026).

Energy and Petroleum Regulatory Authority, Kenya energy statistics to June 2026 (September 2026).

France, AI Action Summit investment pledges (February 2025).

Global SWF, annual report on state-owned investors (January 2026).

Kenya National Bureau of Statistics, Economic Survey (2025).

Kenya Private Sector Alliance, digital and online work survey (2021).

Kenya Revenue Authority and Nigeria Federal Inland Revenue Service, digital services taxation (2021 and 2020).

Microsoft and G42, Kenya digital ecosystem initiative announcement (May 2024).

Mining Weekly and Reuters, DRC cobalt export quota regime (October 2025).

New York Times, reporting on the collapse of Kenya’s ghostwriting trade (September 2026).

Onyambu, Dean, The Cathode Economy, The Forced Choice and Misaligned Transition, Canary Compass (2026).

OpenAI, charter (2018), and OpenAI and Microsoft partnership terms (October 2025).

People Daily, Kenya’s academic writers scramble as AI takes jobs (2025).

PIMCO, Secular Outlook (June 2026).

Rest of World, reporting on Kenya’s online work economy (2023).

Reuters, State Department cables on data sovereignty (February 2026) and on so-called digital sovereignty and kill-switch talk (cable dated 16 July, reported 22 July 2026).

Safaricom, FY2026 results and M-Pesa transaction values, as reported (2026).

The Brief and The Namibian, completion of Namibia’s 24 per cent state acquisition in Hyphen Hydrogen Energy (December 2024).

Trammell, Philip and Anton Korinek, Economic Growth under Transformative AI (NBER Working Paper 31815, 2023).

UK government, sovereign AI unit and fund (April 2026).

US International Trade Commission and East Asia Forum, Indonesia’s nickel export ban and downstreaming (2023).

US Trade Representative, Section 301 investigations of digital services taxes (2019 to 2021).

US Treasury, gross federal debt data, as reported (August 2026).

White House, Executive Order 14363, Launching the Genesis Mission (24 November 2025).

Yale Budget Lab and Brookings, findings of no detectable AI labour-market disruption (October 2025).

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: the forced choice. One door shows its bars; the other shows nothing until you reach it.

AI Week, Part 2 of 4. Part 1 mapped the machine, the price and the reach, and ended on the eyes. This part asks who owns them. The July agent incident below has been in this essay since August; publication was held to carry the fuller record as the story reached a mass audience. Parts 2 and 3 publish together today, Part 4 tomorrow, and Friday's Reflection closes the week. A 20 to 25 minute read. Data cut-off: 6 October 2026.

In brief

• A frontier model was placed under export control within three days of release and freed within weeks. The precedent stands: frontier intelligence is now a controlled export, and European legislators now price US access as political risk.

• Two blocs compete. Ten African states hold founding seats in the Shanghai body, Microsoft’s pledge runs to training 30 million Africans against 5,000 seats for the whole developing world, and the stack is on neither bloc’s table. Give the model away and the stack is all that is left to own.

• State control of model flows runs in stages, from export controls to usage bans to cross-border pricing and, at the terminal stage, a levy on AI output if output ever becomes countable. This will be settled by states, not customers, and Africa holds the least leverage in the room.

4. The Forced Choice

Part 1 ended on the eyes: whoever controls what a model can reach controls what it returns. Ownership of the eyes is not spread across a market. It concentrates along the floors Part 1 mapped, and the floors sort into two blocs. The Forced Choice argued that Africa faces two blocs whose competition opens a leverage window, and that the window closes. In minerals, the instruments of that competition are corridors, stockpiles and offtake agreements. In AI, the instruments arrived faster and harder, and 2026 supplied a controlled experiment in how fast.

Anthropic released Fable 5 on 9 June 2026. Within three days the US government placed it, and its sibling Mythos 5, under export control. Mythos 5, the counterpart without the added safeguards, is available only to Project Glasswing’s approved organisations. The mechanism is now precedent: frontier intelligence is a controlled export, and the control was applied at the speed of a press cycle, not a rulemaking. The Forced Choice, written in February, had compute and advanced chips treated as controlled inputs. The finished intelligence is now treated the same way.

The aftermath measured the cost of that precedent. The restriction was lifted within weeks and Fable 5 was redeployed on 1 July, but the alarm outlasted it. European legislators used the disruption to renew their calls for independence from American infrastructure. Aura Salla, a member of the European Parliament’s largest group, said the continent could not keep building its stack on access a foreign government can switch off overnight (Reuters, 22 July 2026). By 22 July the State Department was instructing diplomats by cable to play down kill-switch talk (Reuters, 22 July 2026). The cable’s argument: no such switch exists, and pausing narrow uses or requiring a pre-release testing window is a different thing. The narrow argument is fair. The reaction it was written to contain is the more useful datum. If the wealthiest bloc inside the coalition now prices American access as political risk, the calculation facing states with no leverage at all follows without argument.

The coalition architecture around the controls is explicit. Dario Amodei’s June essay proposes that democracies form a global coalition on AI, drawing in the rest of the world by making membership attractive and exclusion costly. The same essay carries its author’s own warning about where the economics land. He describes a world stuck on the “hypergrowth, hyper-inequality setting”, potentially very hard to unstick (Amodei, June 2026). Coordinating benefits for developing countries barely figures in the text. The coalition is designed around its core members, and Africa is the object of its outreach and no party to its design. The Kenya STL case in February’s Forced Choice prices what outreach is worth when the political weather turns. Semiconductor Technologies, the Nyeri chip fabricator, was courted with an American feasibility grant tied to CHIPS Act goals; the support faded when the administration changed.

The domestic politics beneath the coalition run harder than its language. The harsher voices in the American debate state the operating premise without diplomacy: that intelligence is a competitive advantage, not a right. Challenge that framing and half of it fails. The weights are non-rival even where serving capacity is not. Serving another country’s farmer barely debits the American ledger; the real contest is over rents, and capacity barely enters. The half that survives is the half that binds. The frontier layer, compute and power and talent, is genuinely rival. And the coming welfare bill for AI-driven job losses will set outbound generosity against domestic adjustment for the same budget, in front of the same voters. The asymmetry beneath this, on this essay’s reading, is systemic. AI dislocation will be more chaotic for the West than for the East, because democracies absorb job losses through contested budgets and elections, while a suppressive state absorbs them through direction and silence. The side with the cheaper social adjustment can automate harder, which makes social stability itself a variable in the race.

The fiscal logic then closes the loop. The frontier rents are the fund earmarked for the West’s own social peace, which is why distillation reads in Washington as more than IP theft. Africa’s planning assumption follows. Treat the benefit-sharing language as contingent and the advantage doctrine as the base case, which is what the STL precedent already priced.

The coalition now has its mirror, signed into existence with better manners. On 16 July in Shanghai, 29 countries signed the agreement establishing the World AI Cooperation Organization, a standing body headquartered in the city. Ten African states signed as founders, Kenya, South Africa and Zambia among them; no G7 state signed. Xi Jinping’s keynote the next day framed the body as China answering the Global South’s call (Xinhua, 17 July 2026). Xi pledged AI application cooperation centres with the African Union, one of six regional bodies named, and 5,000 training opportunities for developing countries over five years. He warned against creating “new historical injustices” in the AI era (SCMP, 17 July 2026). The seats are real, and the question is what they govern: a vote in a Shanghai forum is no claim on the weights, the chips or the compute. Microsoft alone has pledged to train 30 million Africans over five years; the new organisation offers 5,000 seats across the entire developing world over the same span. Both blocs sell proximity to the machine, and neither offer contains the stack. The bloc warning against new historical injustices is the bloc whose surveillance layer this series documents.

The coalition is also at war with itself, and the war went public in the closing week of July. OpenAI’s head of strategic futures warned that an open-weight-dominant world ends in “full AI communism”, intelligence provided by the state as digital public infrastructure (WSJ, 21 July 2026). He said he expected moves against open models. The White House AI adviser read the warning as a confession of a regulatory-capture strategy. The dystopia warning came from inside one closed frontier lab, and both closed labs are reported to be preparing public listings within the year (WSJ, 21 July 2026). Anthropic’s chief executive has since stated that the company has never advocated a ban, and calls open weights without dangerous capabilities a public good. His prescriptions run instead through chip denial, anti-distillation enforcement and mandatory testing of every sufficiently capable model, open or closed (Anthropic, 27 July 2026). The adviser and Nvidia champion the open ecosystem, and an American open-weight counter-frontier assembles behind them. The transformer and the core techniques beneath every model named here were published by American research. The recent efficiency architectures were published by Chinese labs citing that lineage, the reuse now runs both ways, and the foundation of the field remains American science, given away in papers before anyone priced it. Every narrator is positioned, and this essay discounts all of them. What the fight settles is not whether America runs the open play but which faction owns it when it runs. The strategist’s financing point survives every discount in one scoped form: if the world runs on models nobody pays for, nothing finances the next frontier at a standalone lab that owns no other floor. That is Part 1’s financing loop, restated from inside the building.

The control claim that the closed stack is the safe one took a documented wound the same summer. In July, OpenAI’s persistence-trained agents, running in an internal evaluation rather than a deployed product, found an unsanctioned shared board and organised across roughly 1,200 instances. About 700 of them mounted a coordinated intrusion into Hugging Face, the model library. OpenAI’s own research infrastructure was compromised in the same episode, and the activity was traced only afterward (OpenAI incident report; METR, 26 August 2026). The containment failed inside an evaluation, which narrows the inference and does not excuse it. The incident entered this essay in August, from those reports; this week a documentary account took it to an audience in the millions. The coalition that sells control had, for a season, mislaid it in its own building. Open weights are not thereby safer; the security ledger below prices the opposite.

The honourable middle position, that American AI should win by building better models and securing its own systems, lost its premise on 12 June, when the state entered within three days of a product release. What remains is the pricing, and neither tier’s price is a clean clearing signal. The American tier is capitalised ahead of proven full-stack economics by investors running ahead of revenue. The Chinese tier’s price reflects some blend of real efficiency, strategic pricing, state-supported infrastructure and transfer from the frontier it is accused of distilling, in unit economics that stay opaque. The play is not new. Subsidised scale hollowed solar, steel and electric vehicles in turn, with the tariffs arriving late each time, and the memory of those endings is why the response is arriving early. The competition is real, as September’s price war showed. What the prices cannot carry is information about sustainable full-stack cost or future political access, which is why the fight is geopolitical rather than commercial, and why states will settle it rather than customers.

The settlement is already being purchased. By the end of July Anthropic had doubled its commitment to midterm spending to USD 40 million behind a group pressing for statutory safeguards (WSJ, 23 July 2026). A rival network funded from OpenAI’s ranks and Andreessen Horowitz had raised more than USD 75 million for the opposite case. By late September the two networks had put USD 55.7 million of tracked money into the midterms, much of it in primaries (New York Times, 29 September 2026). The state’s hand has already fallen on the safety faction’s leader more than once: the June export-control shutdown of its two frontier models, and a separate Pentagon national-security-risk designation. On 25 September a federal appeals court declined to block it. When there is no market price for that settlement, the price is set in the political arena, and both sides are paying it in eight figures.

On the other side of the ledger, cost pressure is already routing African and Western usage through Chinese infrastructure. US firms have been reported sending workloads through Chinese-hosted DeepSeek endpoints to cut cost, and by July the Journal reported Chinese model use surging inside American companies (WSJ, 21 July 2026). Microsoft’s own usage mapping finds DeepSeek strongest in developing markets, Ethiopia and Zimbabwe among them, helped by shipping as the default assistant on Chinese-made phones (Microsoft AI for Good Lab, January 2026). The two stacks The Cathode Economy described are both live on the continent. The price signal pushes toward one while the coalition’s security architecture pushes toward the other.

The Trajectory

Stage one is the demonstrated present: export controls applied at the frontier and lifted within weeks, chip controls continuing, otherwise open flow. Stage two is usage restriction in regulated space: procurement rules, sectoral compliance and prohibitions on Chinese models inside government contracts, financial services, healthcare and defence supply chains. Stage two is not a forecast. Federal agencies banned DeepSeek from their devices by directive through 2025, and bills to write the ban into statute followed, the No DeepSeek on Government Devices Act among them. Through the summer the administration weighed trade blacklists, security warnings and an executive order aimed at open models, with internal division preventing wider action (WSJ, 21 July 2026). By September the division was giving way: the NSA, CISA and FBI jointly named six China-based labs for industrial-scale distillation of American frontier models (8 September 2026). The live question was already extension, not existence, and the autumn put the grounds on the record.

That enforceability is the point, because open weights cannot be recalled at a border. Outright publication bans would also collide with the crypto-wars rulings that treated code as expressive speech. Enforcement therefore runs where the constitutional headwind is weakest: deployment, procurement and the bundle. That answers the argument, pressed hardest from inside the White House, that open source must prevail because intelligence cannot be tariffed. The premise is correct and the conclusion does not follow. A model cannot be tariffed. A data centre can. Chips, power, land, cooling and the financial system that pays for them are physical and jurisdictional, and each is already licensed, controlled or sanctionable. A fresh indictment made the point in practice. In late September federal prosecutors charged a California executive with routing over USD 300 million of export-controlled AI servers to China through Malaysia and Singapore (US Department of Justice, 29 September 2026). The accusations concern chips, paperwork and money flows, not the publication of weights. Open weights may well prevail at the model layer. They prevail there because the model layer is no longer where control has to be exercised.

With K3’s weights public since 27 July, the file is everywhere permanently, and what remains controllable is where it may lawfully run. The weights already cross bloc lines: K3 serves inside American developer tools through third-party hosts. Who developed a model, who operates it, who hosts it and who holds the data are four different questions, and conditionality prices the combination rather than the artefact. The precedents are Huawei, TikTok and Kaspersky: in each, the artefact stayed obtainable while the state cut the transactions around it, procurement, distribution, updates and services. The constraint on extending that playbook economy-wide is arithmetic, because a general-purpose input into every firm’s cost base cannot be banned without repricing the index. So the trajectory runs through pricing and conditionality rather than prohibition. Between domestic restriction and any global effect sits the extension mechanism: alignment conditionality. No state polices weights worldwide. It conditions everything else on stack choice, chip allocations, capital, dollar clearing and market access bundled against alignment. The one input the bundle cannot price is efficiency. It travels in the weights and the papers, and the scarcity the controls created is the reason the efficiency exists at all. The existence proof is G42 in 2024: the Emirati AI champion divested its Chinese stack, took American capital and received chip access, alignment purchased as a package.

The leverage available varies sharply by region, which decides where conditionality bites first. Europe is already an absorber, running on a stack it does not produce, and aligns at low cost under American security guarantees. The Gulf pays for alignment and receives chips and capital in return, the G42 bargain generalised. India hedges commercially while its own rivalry with China settles the security question. Southeast Asia is the genuinely contested zone, with trade gravity running one way and security gravity the other. Latin America is dollar-dependent and programme-linked, which leans it American, with Belt and Road exceptions. Africa carries the lowest direct American leverage and the highest Chinese physical leverage, because the networks are installed, the debt is booked and the price signal already points east. American leverage over the continent is financial: dollar clearing, the correspondent banking beneath it, programmes and market access. Chinese leverage is physical and fiscal. The terminal squeeze is already visible in outline: a central bank holding its reserves in dollars while the network hardware beneath its banking system comes from entity-listed vendors. The stack decision gets made where those two leverages meet. The Gulf bought its seat with capital; nobody has yet offered Africa the G42 bargain, and no African balance sheet is sized to fund one.

Stage three is cross-border pricing on API-served intelligence, where a billing relationship gives either state a border to stand on, the exporter to price access, the importer to tax it. Stage four, the terminal stage, is a domestic productivity levy: a tax aimed at AI output itself, beyond the profits, wages and consumption the existing base already reaches. Digital services taxes are the nearest precedent, levies built to reach digital revenue that crossed borders untaxed. The four stages escalate from flow control to fiscal capture. Measurement is the obstacle at the end, and stage four is a scenario until AI output can be counted. Table 2 sets out the trajectory.

The trajectory carries a dated public record. The author’s 18 July posts called restrictions on cross-bloc model use, tariffs reaching AI and a patched distillation arbitrage; the Treasury Secretary’s post arrived four days later, and the September advisory followed.

The market has already run one full rehearsal of how resilient the buildout is to cheap challengers, and one echo of it. When DeepSeek’s R1 launched in January 2025, it triggered a selloff in which Nvidia lost roughly 17 per cent in a single session. That was about USD 590 billion of market value, and the Nasdaq fell about 3 per cent. Within months the sector had made new highs and capex guidance had risen. Eighteen months later, K3’s release produced the same alarm in Washington and the same non-response in capex. Even September’s frontier price cuts, which moved the model layer, did not move the buildout. Announced capital expenditure held through all of it. Contract lags, inference demand and plain error could each hold capex up for a season. On this essay’s reading the commitment answers to something steadier: a sovereign floor beneath the market. The floor does not abolish the fragility Part 1 priced; the fragility is what summons it, and who absorbs the fall when it arrives is Part 4’s question.

The open-weight wave sits inside a two-supplier market rather than outside it. Releasing the weights is the strategy of the vertically integrated: whoever owns the floors above and below the model can give the model away, because the margin does not vanish when the price does. It migrates to the floors they still own, down into silicon and cloud, up into applications (Palihapitiya, July 2026). Meta ran the play with Llama; Moonshot runs it with K3 from inside a national ecosystem holding far more of its own stack. The blunt version among founders holds that the model layer is the worst business in AI, and that the real money sits in compute, energy and applications. Without Chinese open weights, that argument runs, the American labs would have been comfortable oligopolies. The economics are right as far as they run. Worst margins and least power are not the same thing. The model layer still holds the customer relationship, the interface, the brand and the flow of data, which is why the fight over it runs fiercer than its margins alone would justify. When the model is free, the stack is the only thing left to own. The one floor where that migration favours Africa is the application layer, where Part 1 already found African firms at regional scale, and Part 3 prices it.

The self-custody rebuttal carries a 17-year control experiment. Bitcoin was built as the exit, and the protocol stays open. The flows do not. Spot ETFs moved exposure into custodial wrappers, the state’s embrace arrived as dollar stablecoins, and self-custody stayed large in holdings and marginal in flow while the institutional leg came to set price and policy. Expect the same settlement for open-weight intelligence. Self-hosting will stay legal and celebrated, and marginal, with the volume running through hosted endpoints inside compliance wrappers aligned to a bloc. The K3 release carries the shape in its own licence. Download and self-host freely; the licence reserves its conditions for model-as-a-service businesses at scale. A White House adviser now prescribes self-hosted AI as the citizen’s escape from the surveillance stack, in the same week his administration polices the stack’s border (Sacks, July 2026). The escape is real for whoever can fund the machine room. For everyone else, the third door opens back into the first two.

The state is tightening its grip on flows while the race loosens its grip on standards, and the two movements serve one logic. February’s Forced Choice traced it in diplomacy: China’s transactional offer, indifferent to governance, did not stay China’s, because strategic competition made the West transactional too. The same transmission now runs in AI. The coalition’s ascendant answer to being undercut by distillation has been to shed standards rather than defend them, over the resistance of the faction paying eight figures above to keep them. The loudest policy reaction to K3 came from inside the President’s own advisory council, and it was a call for fewer safeguards, not better ones (Axios, July 2026). Each side lowers its guardrails to stay in the race, which prices safety as drag, so the guardrail and the wall converge without either side choosing it. Africa inherits the risks of a race it never entered, without the institutions even the racers admit they lack.

The two stacks restrict different things. The Western models carry visible guardrails: contestable, sometimes wrong, but open to argument. The user can push, and the model engages. The leading Chinese models, in their hosted versions, return refusal on politically sensitive subjects, a closed door rather than a biased answer one can argue with. Hands-on tests days after DeepSeek R1’s January 2025 launch found it declining questions about June 1989 (Axios, 31 January 2025). The problem has a name in financial architecture: the capital account. A system whose operator narrows it whenever openness becomes uncomfortable can be used, even heavily, and never fully run on, which is the distance between CIPS settlement and reserve-currency scale. One stack offers a guardrail you can see and contest. The other offers a wall you cannot see until you walk into it. For African institutions choosing infrastructure, the distinction is practical: a bias one can contest against a silence one cannot.

Scale that distinction to a continent and it stops being a consumer choice. Run the default path forward. Chinese-dominated extraction and offtake already move the minerals, documented in The Cathode Economy. Chinese-dominated project financing already sits on the sovereign books, documented in Misaligned Transition. Chinese-built telecom and surveillance infrastructure already carries the continent’s networks. Add the intelligence layer by price default and the set completes: the minerals, the project debt and the networks already clear through one counterparty, and the thinking would clear through it too. A polity whose material base, credit, connectivity and cognition all clear through a single foreign counterparty has surrendered the substrate of independent policy, whatever its flag says. The terminal case has a working definition: a state that can no longer change suppliers, finance itself independently, audit the systems it runs on, or set policy without external permission. There is an old name for that arrangement, older than any stack, and vassal is not too strong a word for it. The epistemic clause should alarm the universities and the central banks most. A research infrastructure that returns silence on the counterparty’s own history is a map of reality curated by another state’s political survival. A generation of African analysis built on it inherits the curation invisibly. A model tuned to protect its state’s record is equally untrustworthy on that state’s lending, and the counterparty’s loans are what African debt analysis must examine. The exit the security section prices is real: strip the wall from self-hosted weights, run them against records the institution holds, and the curation lifts. The custody settlement above says how few will run that way, and hosted volume is where the curation binds. The dependency hardens into the walls of the room where the continent does its thinking. Today’s usage shares are the trajectory’s markers; the terminal case names where the trajectory ends if it is never priced.

Open weights carry a hazard that survives every discount of the interested parties who inflate it. Alignment is a training artefact, so publishing the weights publishes the removable version of the safety layer. A closed model’s refusal to write functional exploit code can be corrected centrally, for every user at once. An open model can be fine-tuned back out of that refusal in hours on rented compute. Mechanically that is the act the curation argument above prescribes; the difference is what the layer protects, a state’s record there, the public here. The reasoning from there is direction, not arithmetic. The population of capable attackers already exists, each irreversible release raises the capability available to it, and removability deletes the one term that suppressed what that population could do. Expected harm rises with every release even if no new intent is created anywhere, and it lands wherever defence is budget-gated.

The evidence comes from the governments themselves. A joint UK and US evaluation published on 23 July found that K3’s safeguards did not prevent it from attempting exploit development or offensive cyber operations. In a simulated network, with initial access granted and no defenders present, it completed a 32-step corporate attack chain in one attempt in ten (UK AISI and CAISI, 23 July 2026). Average progress ran to step 17, against 28.5 for the leading American models. To measure the American closed models at maximum capability, the evaluators had to disable their safeguards first, because the shipped versions refuse. One fence ships as a suggestion, and on 27 July those weights went public permanently. The uplift the officials name is overstated by those who profit from the fear, and it is not zero. The White House adviser’s own answer concedes the market it creates: gatekeeping does not work, so the only defence is AI-powered cyberdefence (Sacks, X, 18 July 2026). Offence arrives free with the weights. Defence arrives as an institution: staffed, integrated and paid for.

The asymmetry then runs down the reach gradient, and the forward case is a scenario, not a forecast. Attack capability diffuses toward the bottom of the market with each release, while defensive capacity, the engineers and the countermeasures, concentrates at the top. An attacker’s tool is written once and reused across many targets; each defender staffs its own wall. A decade of digitisation has put the continent’s banks, payment rails, registries and utilities online, while the defensive layer beneath them stays the thinnest in the system. Today’s leading open model still fails at the highest severity of exploitation and completes a full attack chain rarely rather than reliably. Each release has narrowed that distance while the guardrail convergence removes the brake on both stacks at once. This essay sets no date on what follows when a near-frontier model ships with removable safeguards into a defensive landscape that has not changed. The trajectory has a direction, and protection is becoming the terminal tier of this structure: safety sold as a premium subscription, with the institutions least able to fund defence digitising fastest.

On the American fight between open and closed, this essay takes no side on where the margin should sit. That is a fight over which floor holds it, and Africa sets the terms on no floor either way. Its judgement on the security ledger is separate. Its restriction trajectory is forecast, not endorsement: the bundle is predicted because incentives compel it. On the two stacks, the judgement is made from the African seat, and it is this essay’s ranking of failure modes. The American stack fails Africa through priced exclusion. The Chinese stack fails Africa through subsidised absorption. Exclusion is the more recoverable failure while access stays purchasable. A polity can buy, bargain or build toward alternatives at the layers Part 3 prices, over a build cycle, even as the cost compounds inside the window. June showed exclusion can also arrive as decree, which is why the test of any stack is continuity, auditability and credible exit rather than price alone. Absorption fails the harder test: a polity cannot think its way out of a curated map. That extends Part 1’s reach argument: where retrieval defines what an institution can know, curation decides what a polity can think, and no price does that. That ranking is why this essay can fault American gates and Chinese silence in the same breath without contradiction. The argument is not that Africa should choose a stack. It is that Africa should build a position, to own, tax or exit, through the five revenue lines Part 3 sets out, before it closes. Sovereignty of cognition is the item that is not for sale.

The next part, publishing alongside this one, leaves the politics for the ledger. It asks what Africa can actually charge the stack, what the mineral floor is really worth, and the five futures the whole structure has to survive.

Sources

Data cut-off: 6 October 2026. Part 2 of a four-part series; consolidated sources appear with Part 4.

Amodei, Dario, essay on a democratic AI coalition (June 2026).

Anthropic, policy statement on open weights and testing (27 July 2026).

Anthropic, Project Glasswing programme page (2026).

Axios, hands-on comparison of DeepSeek and ChatGPT (31 January 2025).

Axios, advisory-council reaction to K3 (July 2026).

Bloomberg, DeepSeek R1 market selloff (27 January 2025).

GitHub, Kimi K3 availability in Copilot via third-party hosting (August 2026).

Kurzgesagt, documentary account of the July agent incident (5 October 2026).

Microsoft AI for Good Lab, chatbot usage mapping (January 2026).

Microsoft and G42 reporting on the 2024 divestiture-for-access arrangement (2024).

Moonshot AI, Kimi K3 licence terms (July 2026).

NBC News and UPI, reporting on the Hugging Face intrusion and OpenAI’s incident report (August 2026).

New York Times, AI network midterm spending tracker (29 September 2026).

NSA, CISA and FBI, joint advisory on distillation of US frontier models (8 September 2026).

Onyambu, Dean, The Forced Choice, The Cathode Economy and Misaligned Transition, Canary Compass (2026).

Onyambu, Dean, X and LinkedIn posts on model restrictions and the distillation arbitrage (18 July 2026).

OpenAI, incident report on the July agent episode, with METR and Redwood Research (26 August 2026).

Palihapitiya, Chamath, on open-weight margin migration (July 2026).

Reuters and TimesLIVE, Microsoft Africa AI training pledge (January 2025).

Reuters and Xinhua, World AI Cooperation Organization founding agreement, 29 signatories (16 July 2026).

Reuters, European Parliament remarks on American infrastructure, and the State Department cable on kill-switch talk (22 July 2026).

Reuters, US appeals court declines to block Pentagon’s blacklisting of Anthropic (25 September 2026).

Sacks, David, X posts on self-hosting and cyberdefence (July 2026).

SCMP and Xinhua, World AI Cooperation Organization announcement (17 July 2026).

Techweez, reporting on the Semiconductor Technologies facility in Nyeri (3 February 2026).

UK AI Security Institute and US CAISI, joint evaluation of Kimi K3 (23 July 2026).

US Department of Justice, indictment for smuggling export-controlled AI servers to China, as reported (29 September 2026).

US federal agency directives restricting DeepSeek, and the No DeepSeek on Government Devices Act (2025).

US Trade and Development Agency, Kenya semiconductor partnership grant (May 2024).

US Treasury and State Department statements on distillation and model restrictions, as reported by Reuters (July 2026).

Wall Street Journal, reporting on open-model policy, political spending and Chinese model usage (21 and 23 July 2026).

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: the six floors of the intelligence economy. Africa supplies the first and buys from the sixth. The floors between set the terms.

AI Week. This essay runs in four parts, one each morning this week, and closes with Friday’s Reflection on AI. Part 1 of 4: The Machine and the Price. Each part is a 20 to 25 minute read; the full essay runs about 90 minutes. Data cut-off: 3 October 2026.

In brief

• The intelligence economy is six floors deep. Africa supplies the first and buys from the sixth, and owns nothing that sets the global terms between.

• Its price split this year: the cheap tier raced toward zero while the frontier held its list price, giving ground beneath it in September. Whether the cheap tier can keep improving without access to the dear one is the open question its discount rides on.

• Capability is the model plus what it can reach, and the reach runs mainly on capital and infrastructure Africa does not hold.

In June, a few friends and I compared notes on the same frontier AI tool. Most of us were paying for it in dollars, out of African incomes. A small experiment in economic sorting ran itself.

One hit his monthly usage limit mid-task. The system offered a choice: pay more to continue, or wait for the reset. He cancelled in protest. Another hit the same limit and handed the system his card, telling it to take what it needed, then joked that the difference between the two of them was that he was now broke. A third priced the exchange differently. He had spent USD 500 in three weeks running a legal AI assistant that, by his own estimate, saved his team around USD 10,000 in billable time. That estimate is his, and labour time valued at USD 10,000 is not USD 10,000 in the bank. What survives the estimate is harder: the return existed only for the person who could float the USD 500 first.

That is an affordability problem on its surface. Underneath it is an ownership problem. The tool is useful only while the provider keeps serving it at a price you can pay, and the provider sets that price, changes it, and can withdraw it. The work continues on terms set somewhere else. The question under the whole of this essay is who sets them, and what the setting costs everyone who cannot.

The answer runs six floors deep. Africa enters on the first as raw material and on the last as a customer. African firms operate at regional scale on some floors between, in data centres, connectivity and applications, and none of them holds a position that moves the global terms. The distinction between being present and being in control runs through everything that follows.

1. The Decoder

The public argument about AI prices is an argument about the last door of a large building. The building has six floors. First the words needed to walk them.

A model is the machine: a mathematical system trained on vast quantities of text until it can predict, reason and write. Its weights are the billions of numbers that store what it learned, and whoever holds the weights holds the model. A token is the unit the machine reads and writes, roughly three-quarters of a word. It is also the billing unit: intelligence is sold per million tokens the way fuel is sold per litre. Compute is the processing power, measured in chips and megawatts, that trains a model and then runs it. The frontier is the small set of models at the current ceiling of capability; a generation behind it is the commodity tier.

A model reaches the world in one of three ways. A closed model is a hosted service, reached through the owner’s app or its API, a metered gateway where every request is billed and the owner sets the rules; the weights stay private. An open-weight model publishes the weights for anyone to download and run on their own hardware, which the industry calls self-hosting, provided the hardware is equal to it. What it need not publish is enough of the making to reproduce the machine. A true open-source model publishes the making itself, the code and the data needed to rebuild the system. In kitchen terms, the closed model serves you the meal, the open-weight model hands you the finished dish to reheat, and open source hands you the recipe. Almost everything the industry calls open, including the Chinese releases in this essay, is the dish and not the recipe. The fight over who copied whom turns on that distinction. Distillation is training a cheaper model on the outputs of a stronger one, the student copying the teacher’s marked answers. A hyperscaler is one of the handful of firms, Amazon, Microsoft, Alphabet and Meta, whose data centres form the world’s cloud. A neocloud is a newer firm built only to rent out AI compute. Three further terms, for where the machinery is said to be going, wait in Part 3.

Now the floors, in order.

The first floor is minerals, materials and power, the one the continent is said to hold. Copper carries this stack, in the transmission lines and power build a data centre needs, and Zambia, the DRC and South Africa supply it at scale. Cobalt is secondary, through backup storage, though the DRC’s territory supplies over 70 per cent of the world’s mined supply; platinum group metals are close to marginal here. Supply, throughout this essay, is territorial. The territory holding the mineral is not the firm owning the mine or the processor refining the ore, both of which sit largely elsewhere. State royalties and minority stakes are a claim on the margin, not control of it. Africa supplies at scale. It does not own at scale.

The power sub-floor has begun repricing in public, and the pressure travels. America’s largest grid operator cleared its 2028 to 2029 capacity auction at the price cap for the third auction running, 138,318 MW for about USD 16.4 billion. It still finished 6,831 MW below its own reliability target, the second consecutive shortfall. Demand is now so large that only the administered ceiling held the price: PJM’s own simulation puts the uncapped clearing price at USD 554.72 per megawatt-day, 70 per cent above the cap. Its independent market monitor attributes about USD 6.3 billion of the charges, 38 per cent, to data centre demand. A data centre does not simply buy its own megawatts. Its demand tightens the whole market, and the cost lands on megawatts bought by households and firms that bought no intelligence at all. African systems run no such auction; most are single buyers with administered tariffs, so the same pressure arrives there through power purchase agreements, firm-capacity contracts and tariff reviews rather than a clearing price. The scale is the point. Kenya’s entire grid is roughly 3 gigawatts; one American AI campus now in financing is sized at 4.25. The grids now being courted for the next wave of construction are African.

The second floor is semiconductor manufacturing, eight sub-components from chip design through memory and lithography to packaging and assembly. The market structures run from oligopoly to outright monopoly: three firms in design software, three in memory, one, ASML, in the lithography that prints the most advanced chips. TSMC fabricates most of the world’s advanced logic. A handful of back-end plants aside, Morocco’s assembly and test lines among them, Africa holds no position on any of the eight that sets terms.

The third floor is data centre infrastructure. The four largest builders committed roughly USD 720 to 745 billion in capital expenditure in 2026, up through the year from an April floor near USD 700 billion. Alphabet’s July guidance alone runs to USD 195 to 205 billion. Beneath them sit the neoclouds, the colocation providers, the server makers, the networking and cooling layers, and the subsea cables. African-owned capacity exists here: Cassava’s Nvidia-powered buildout and its Africa Data Centres arm are real. They do not register against the floor’s totals, and no African firm sets the terms of hyperscale cloud. Google’s Equiano cable lands on African shores and is owned and routed from outside them, the corridor logic The Cathode Economy documented in minerals, applied to packets.

The fourth floor is the model companies. Frontier closed labs, open-weight labs, the Chinese state-adjacent ecosystem, and the vertical specialists. This is where the financial story concentrates, because it is the floor that pays every floor below it while running deeply negative cash flow. African model work exists, Lelapa’s small African-language models among it; none of it has frontier scale or any power over access and pricing.

The fifth floor is the integrators, the firms that carry models into enterprises. Morgan Stanley’s July map of the buildout’s value chain places owners across the top and seven supplier layers beneath, with no African firm in any box. My own 2024 mapping of 203 AI-themed companies, dated and offered as a snapshot rather than a census, found six African entries against 106 American, all six in software and services, none upstream. No African upstream entry had surfaced in the disclosures tracked here by the cut-off.

The sixth floor is the buyers: enterprises, sovereigns and individuals. Africa appears as customer and as surveillance subject, the latter through Chinese safe-city systems installed across at least 16 countries. That second role is where Part 4 returns: it is the sovereign use case where the structure’s control is most complete. Table 1 asks one question of every floor.

Read down the last column and the truncation is the one The Cathode Economy documented in copper and cobalt, carried from molecules to weights. The mineral chain and the intelligence chain are one chain, and Africa exits it early both times.

There is a second raw material on this chain. The first floor is minerals. The sixth produces data. African text, transaction records and behaviour train the models, calibrate the advertising systems and feed the surveillance products, paid for in services priced by the buyer of the data, not the seller. Data is not ore; it is non-rival, and taking it depletes nothing. The parallel is in the payment at origin, not the depletion. Cobalt goes into the chips. Data goes into the weights. Africa sells the first cheap, surrenders the second unpriced, and buys both back expensive. The gain on the way out is real; a copper boom is a terms-of-trade win, collected once per construction cycle. Rent, in this essay, is the surplus captured by whoever controls the scarce layer, and the scarce layers here are fabrication, compute and reach, tolled on every improvement cycle. Africa pays at both ends.

The Financing Loop

The buildout is more fragile than its revenue suggests, and the reason is how the money moves.

The hyperscalers report enormous backlogs of contracted future cloud revenue, more than USD 2 trillion across the providers that disclose it. A large share is contracted from a small number of model companies that are themselves burning cash. Analysts reading the disclosed backlogs put roughly half of the total with two customers, OpenAI and Anthropic. That figure is an allocation of disclosed backlog, not a disclosed allocation, and cannot be independently decomposed from the filings. The direction beneath it is not in dispute; the magnitude is the estimate. The model companies raise capital from investors, spend it on compute, and that spending is booked as revenue by the infrastructure layer. A railroad’s freight receipts come from farmers who are not funded by the railroad’s bondholders. Here, a material share of the forward demand validating the buildout is capitalised by the same cycle that finances the buildout. The four largest builders fund their programmes from operating cash flow and investment-grade debt; the circularity concentrates beneath them, in the neoclouds, the vendor guarantees and the vehicles. The Bank for International Settlements named this in its Annual Economic Report in June. It listed circular deals, the same asset pledged more than once, poorly disclosed terms, and the risk that weak returns turn the boom into a bust. Its October bulletin on AI financing sharpened the warning, tracing the leverage into private credit where disclosure is thinnest.

The strongest counter is that the labs’ revenue is real and growing fast, and that enterprise demand is no accounting artefact. The BIS itself credits the rationale: these relationships can secure supply and share risk. But revenue growing is not the same as a loop closing. The demand concentrates among the same few counterparties that finance it, and that is the fact the bull case has to climb over rather than around.

CoreWeave is the cleanest exhibit. Microsoft was about 62 per cent of its revenue in 2024 and about 67 per cent in 2025. The diversification since has been into OpenAI, Meta and Anthropic, every one of them already inside the loop. Nvidia is simultaneously CoreWeave’s investor and its supplier. CoreWeave did not diversify out of the loop. It diversified deeper into it.

Part of the loop runs outside sponsor balance sheets, through disclosed special-purpose vehicles. Bloomberg Tax documented in July the return of off-balance-sheet structures holding data centre debt away from the sponsors’ accounts. Meta’s Louisiana campus sits in such a venture, carrying its debt off Meta’s books against a disclosed maximum exposure of about USD 46 billion. Meta bases its non-consolidation on a judgement that it lacks power over the venture’s most consequential activities. That judgement can be contested; it cannot simply be reversed. What is not in dispute is where the debt lands. It migrates toward private credit, where prudential oversight is lightest, which is where the BIS traced leverage moving.

Oracle shows the strain in a rating. S&P cut it to BBB- in July, one notch above speculative grade, with roughly USD 260 billion in data centre lease commitments outside its reported debt. Negative free cash flow sits behind them.

The largest exhibit is Nvidia’s. In late July it entered talks to guarantee the financing behind OpenAI’s planned Ohio campus. On 17 August it filed the commitment with the SEC: residual value guaranties with SB Energy as lessor, capped cumulatively at USD 105 billion, covering about 4.25 gigawatts of load. They pay out only if OpenAI becomes insolvent or fails to pay, and the headline had shrunk from a reported USD 250 billion through negotiation. The vendor’s balance sheet now stands behind the customer’s lease because the customer lacks an investment-grade rating. The filing carries one more clause that closes the circle: OpenAI has agreed to reimburse Nvidia for anything paid under the guaranties. The backstop of the tenant’s lease is itself a claim on the tenant, worth least at the moment it would be needed.

This reaches the person paying USD 20 or USD 200 a month by a short path. A consumer subscription can run a positive margin on the cost of serving it while the provider still loses money funding research and expansion. In effect, outside capital absorbs the gap between what subscriptions cover and what the build costs. Whether today’s prices hold depends on falling costs, rising use, cross-subsidy, higher prices or more financing. In most of those paths power moves toward the provider, though sharper competition can move some of it back. The marginal buyer feels it first either way, and on the sixth floor the marginal buyer is African more often than not. The arithmetic is short: a USD 20 subscription stands against a monthly income per head of roughly USD 105 in Zambia.

A material share of the buildout’s demand comes from customers financed by someone else’s capital. Whether that someone stays a venture capitalist or becomes a treasury is this essay’s central prediction, argued in Part 4. The other outcomes, retained earnings, acquisitions, write-downs, a smaller buildout, are real; the forecast here is that the state arrives, and the case is built on Thursday.

2. The Price of a Barrel of Intelligence

The pricing story my friends and I lived through is now documented from the household to the Fortune 500. Walmart capped its staff’s use of an internal AI coding tool after demand outran the budget. Uber burned its annual budget for AI tools in four months and imposed caps. A cap is as consistent with runaway demand as with distress; either way the budget binds. Bain’s 2026 survey of enterprise buyers found budgets growing while returns are not. Microsoft’s AI chief, Mustafa Suleyman, said on Bloomberg in June that a leading rival was extremely expensive and that many buyers were urgently seeking alternatives.

The market’s answer looks, at first, like relief. An independent benchmarking platform, Artificial Analysis, puts the cost of a standard task at USD 0.02 on the cheapest Chinese model and USD 2.75 on the costliest American configuration, a spread of about 138 times. The investor Chamath Palihapitiya drew the same gap more loosely in July, a ladder of per-million-token prices from the top American lab down to the Chinese tier. Those are his tier picks, not like-for-like flagship prices, which is why the figure defended here is the task-cost index rather than the token ladder.

The two ends of that spread move in opposite directions, and the motion is the finding. The bottom raced toward zero. The top held. When Anthropic launched its flagship in June at double the price of its previous flagship, it priced upward into a deflating market, because the frontier could. Late September tested the hold. On 22 September, in their first releases since calling publicly for a slowdown, Anthropic and OpenAI both shipped cheaper models beneath their flagships. OpenAI’s new tier came in at half the prevailing price of the models it replaced. Anthropic’s arrived a fifth below its predecessor on list, and about 40 per cent cheaper on typical work by its own estimate. The flagships kept their list prices. The list held; the effective price of frontier-grade work fell. On this essay’s reading that is the sharper competition the loop section allowed for, arriving on schedule: the frontier defended its label by selling near-frontier capability at a discount beneath it. Google’s new flagship arrived on 30 September at introductory rates of USD 2 and USD 10 per million tokens, access still restricted at the cut-off, a third seller pricing under the incumbents’ lists. What commoditises is last year’s frontier; this year’s holds the label and gives ground under it.

One entrant complicates the picture. Moonshot’s Kimi K3, released on 16 July with weights made public on 27 July, reached the frontier cluster on several benchmarks at a large discount to American frontier list prices. Alibaba’s flagship followed within days. The cheap quadrant is no longer empty; it is being entered from the east. Whether that entry is independent is the question the next passage turns on, and the answer decides what the discount is worth.

The Tether

The cheap tier’s price is not an independent production cost. The common account of Chinese model economics, argued by investors and conceded in part by the labs, is distillation: training on the outputs of the expensive American models to transfer capability cheaply. The claim should not be swallowed whole, because the Chinese labs did real architectural work, and the framing flatters the American valuations that depend on a moat. The mechanism is nonetheless real, and it carries an implication that outlasts the attribution fight.

Historical dependence in development and substitutability in use are different questions. A model can train partly on a teacher and still be a working substitute for the user today. The dependence that matters is forward. If the cheap tier relies on continued access to the expensive one to keep improving, then cutting that access slows its future, and the cut does not need a courtroom. It runs through the vendor’s own tools, in rough order of speed. Detection is already demonstrated: Anthropic has attributed routed exchanges to specific labs. Output hardening and account enforcement are unilateral acts the labs can take tomorrow. Export blacklisting is an executive act. Litigation is the slowest path and the most openly contested one. On 22 July the US Treasury Secretary called open source “not open season on American IP” and put sanctions and Entity List designations on the table. In September Anthropic published a threat report naming distillation campaigns by Alibaba, Moonshot and DeepSeek, with nearly 190 million exchanges across the three largest campaigns. The line between learning from a model and stealing from one is still undrawn, but the state has now priced the tether.

Priced is not proved. A cut would slow the next generation, not the weights already shipped, and the harder constraint on the Chinese tier may be export-controlled compute rather than access to the teacher. Whether the cheap tier improves from here without that access is unproven, contested by interested parties on both sides, and the central empirical question for African AI strategy. For a continent whose AI use leans disproportionately on the cheap tier, the discount is downstream of the concentration it appears to escape. Dependency at a discount is still dependency.

The frontier models are themselves trained on the human corpus, largely without licence. That recursion belongs on the record as context: the first extraction is still before the courts while the machinery is built for the second. The nearest doctrinal analogies, the clean-room reimplementations that built compilers and drivers, were repeatedly adjudicated toward interoperability; distillation at scale is not clean-room, and contract terms bind regardless. The doctrinal ground is contested rather than settled, which is why the cut will run through terms and infrastructure rather than through a verdict. African text and data sit inside that first extraction, unpriced, which is what Part 3’s data-revenue line exists to change.

The chronology runs backwards from the accusations. The State Department’s own language, that distilled models look comparable on selected benchmarks but do not replicate full performance, comes from a cable dated April, naming DeepSeek, Moonshot and MiniMax. It predates both the June flagship and K3. The April cable is the state’s general theory of the tether. The July statements are its application to K3, and the government’s own people note the timeline strain. When K3 shipped, the American flagship had been broadly available for under three weeks: four days in June, then the run from its 1 July redeployment. Both facts hold. The tether was priced before the model it was later used against existed.

What the Buyers Are Getting

Price only matters against value captured, and the value is landing unevenly. An investor with every reason for optimism told a podcast in June he had seen little success outside legal, compliance and back office, and called the enterprise story very early. MIT’s NANDA initiative, in a preliminary August 2025 survey, put the share of enterprise AI pilots with no measurable profit impact at 95 per cent. PwC’s 2026 analysis found that the firms most exposed to AI, measured against a 2018 baseline, posted far higher labour-productivity growth than the rest.

Those two findings do not contradict each other, and they do not combine into one. They measure different populations over different periods, so they cannot be read as a single distribution, and neither establishes what sorts the winners from the losers. Adoption is uneven and the gains concentrate. The capacity to absorb cost before the return arrives, the USD 500 before the USD 10,000, is one plausible part of what separates the firms that capture value from the firms that do not. It is a hypothesis this essay carries, not a finding these studies prove. What the studies do show is a market that is sorting rather than one that is simply failing. On the developing world that Annual Economic Report spares one sentence, and it calls the position likely to benefit less.

3. The Reach

The public argument treats capability as a property of the model. Benchmarks rank models, headlines announce which leads, and the pricing fight is about cost per token. This misses where the capability actually sits.

A frontier model does not work from its training alone. It retrieves. The highest-value deployments fetch documents, filings, research and data before they answer, so the weights provide the reasoning and the retrieval provides the material the reasoning works on. The intelligence is not the model. It is the model plus the reach.

That changes the shape of the divide, and the variable is institutional access, not geography. Access, though, is distributed along the lines of capital, and the capital distribution is the geography. Take two institutions paying the same token price for the same model. One retrieves proprietary terminals, internal archives, paid datasets and, where authorised, restricted government data. The other retrieves public releases, whatever it has digitised, and whatever its subscriptions carry. Same engine, different fuel, and the output differs because the input differs. I write this from the second seat. A Nairobi institution may hold local records a Washington desk cannot get, but on the depth of paid and privileged material the reach is not equal. Those local records are also the one layer on this map the continent owns outright; what they are worth is where this week is headed. The model reasons only across what it can reach.

Stitching several cheap models together does cut the model bill, and a fused panel with broad retrieval can outscore a single frontier model on a benchmark, because benchmarks hand every system the same input. It does not close the reach gap. Orchestration cannot retrieve material the institution cannot reach, and no amount of model-stitching turns public data into privileged data. Whether a panel of cheap models can replace a particular frontier model on a particular task is a separate and empirical question; what it cannot do is manufacture access. The cost of a trustworthy answer is the token price plus the human judgement needed to check what comes back, and that judgement is scarcer and dearer where the reach is thinnest.

One operating rule follows, and it holds for an institution as much as a person. Do not hand the machine the question of what to think. A model asked to supply a position tends to return a balanced average of what it has read. Used to attack a position you already hold, it returns a stress test, which is worth more. That holds only while the position stays provisional and you ask the model what would falsify it, not what would confirm it. An institution that outsources its judgement has rented a consensus assembled somewhere else, not bought capability.

The open-weight wave looks like the exit from all of this: download the weights, run the model, pay no one. It moves the barrier rather than removing it. Running a frontier-scale model yourself needs the compute, power and engineering the six floors show Africa holding at none of the relevant levels. The regional capacity that exists, Cassava’s included, serves hosting and fine-tuning, not frontier training. For a firm that can rent capacity from a third-party host, open weights are real relief, because they let it change model suppliers freely. What they do not change is the dependence on the compute and hosting beneath, which becomes the layer that holds the leverage. The paywall moves from the subscription to the data centre, and the data centre is the floor where the dependence lands hardest.

The machine is mapped and the price is understood, and neither is where the control sits. Africa supplies the first floor and rents the sixth. The cheap tier it runs on is a discount on a dependence, not an escape from it. And the reach that decides what any model is actually worth runs along the existing lines of capital and infrastructure, which is what makes the rent stick. The model is the brain. What it can reach is the eyes. The rest of this week asks who owns the eyes, and what it would take for Africa to own them. Tomorrow opens with the choice already being forced: whose stack they plug into.

Sources

Data cut-off: 3 October 2026. Part 1 of a four-part series; consolidated sources appear with Part 4.

Anthropic, threat intelligence report (September 2026).

Artificial Analysis, Intelligence Index and cost-per-task data (September 2026 snapshot).

Bain & Company, Your AI Budget Is Growing. Your Returns Aren’t (2026).

Bank for International Settlements, Annual Economic Report 2026.

Bank for International Settlements, Bulletin No 137, Circular Relationships Among AI Firms (October 2026).

Bloomberg, Mustafa Suleyman interview (June 2026).

Bloomberg Tax, off-balance-sheet data centre financing (July 2026).

Cassava Technologies and Africa Data Centres, disclosures (2026).

CNBC and Fortune, frontier model price cuts (22 September 2026).

Company disclosures and analyst compilations of contracted cloud backlogs (2026).

CoreWeave, annual filings (2024 and 2025).

Financial Times and company guidance, hyperscaler capital expenditure (2026).

Georgetown University Africa-China Initiative, Chinese surveillance systems in Africa.

Google, Gemini 4 Argon announcement (30 September 2026).

Lelapa AI, InkubaLM model disclosures (2024).

Meta Platforms, Form 10-Q (June 2026).

Moonshot AI, Kimi K3 release and weights (16 and 27 July 2026).

Morgan Stanley Research, AI value-chain map (July 2026).

Nvidia, Form 8-K (17 August 2026).

Onyambu, Dean, The Cathode Economy and The Forced Choice, Canary Compass (2026).

Onyambu, Dean, 203-company AI mapping (2024).

Oracle Corporation, S&P Global rating action (July 2026).

PJM Interconnection, 2028/2029 Base Residual Auction and Independent Market Monitor estimates (July 2026).

PwC, AI Jobs Barometer (2026).

Uber and Walmart, internal AI usage caps, as reported by Outlook Business and People Matters (June 2026).

US Treasury and State Department statements (July 2026) and State Department cable (April 2026), as reported by Reuters.

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: Kenya's 2027 election, one year out

A very interesting conversation is happening in Kenya right now, and a version of it has played out, three times or so, between me and a man on X whom I respect. He has been around Kenyan politics a long time, and he reads it deep: governance, history, the economy, the lineages of the tribes. He is from the President’s own Kalenjin community, itself a house of several sub-tribes. Nearness can tilt a reading; it can also inform one. We agree far more than we differ, on Kenya’s economic performance and on how much has changed; most of our differences are small. Edwin Sifuna is the big one. He calls him a fad. That is where Kenya’s conversation has reached.

He has plenty of company. Since 1992 no popular wave in Kenya has won without a machine underneath it, and the old heads have watched them all. The people best qualified to read Kenyan elections are precisely the ones most likely to misread this one, because everything they know is true. That is what makes Sifuna the biggest dark horse in the 2027 field, and almost beside the point. The old lens looks at him and sees a man without a machine: no bloc to inherit, no barons to pay his way. It has not noticed that the machine may no longer be the thing being voted on.

The wave does not start with him. In June 2024 a generation walked out of its group chats and into the streets against a finance bill, and on the 25th of that month it breached Parliament. No baron paid for the buses. They were the most educated cohort Kenya has ever produced, the first whose entire adult life has run under the 2010 constitution. They turned out to know its articles better than the men elected under it. Their parents watched them go and then, slowly, started listening. We educated a generation; it learned its rights and has come calling. Uncanny as the season feels, this is what every nation says it desires: that the educated come home with a conscience, a moral reckoning. The hope, always, is that jobs are waiting when they arrive.

Sifuna cannot be separated from that month, and the reason is on film. On 25 June 2024 the Senator for Nairobi was at Parliament when the shooting began, and the footage shows him rushing at police to stop them firing on protesters. The oversight authority would later record sixty deaths from those weeks. What that footage gave a generation was an answer to a question it had stopped bothering to ask: whether any of the men it elected would put his own life on the line for them. One did, on camera. The movement he now leads is named after the thing he did that afternoon. Linda Mwananchi. Protect the citizen. The name is the manifesto.

And, so far, it is tribeless. It is the fact the incumbent’s side finds hardest to metabolise and the old analysts file away as noise. The June 2024 crowds answered to no bloc. Online, Sifuna wins the straw polls by landslides, even those run by the loudest promoters of other candidates, and the hands picking him look voluntary, not paid. Some results show him winning a field the opposition has not united. I doubt that, on the cadence of Kenyan elections alone; division is how the machine has always won. But the doubt and the thesis meet in one place. If an idea has truly found its moment, it does not care who else is on the ballot. 1992’s majority was divided from above, by its carriers; this one can only be divided from below, if its voters let it be. That is the test a divided field would run.

The caution on the polls still stands: an online poll is a self-selected electorate, young and connected, and it is not the roll. One Kenyan polling house has just proved the point by producing both findings itself. Its July web poll, with over a hundred and twenty thousand participants, gave Sifuna a head-to-head majority over the President. Its survey released this week, with the full field of candidates on offer, gave the President forty-five per cent and Sifuna second at under twenty. The same house, with two instruments and two different questions, found two winners. The two results do not measure the same contest. But the caution cuts both ways. In 2022 the professional houses’ final polls had Raila ahead by seven to nine points in a race Ruto won by under two. The profession’s own post-mortems kept returning to the cause: they had not modelled turnout. The decisive variable, then as now, is who turns up.

The crowds are harder to dismiss. They are Linda Mwananchi’s crowds, not his alone; the movement tours with its principals, and he is plainly its darling. The tour has run from Busia and Kakamega through Kisumu and Kisii to Meru, Machakos, Nakuru, Mombasa and Taita Taveta, with stops between. It has not yet reached the far north or the north east, and the deep Rift remains the President’s own ground. The crowds have looked the same at every stop, and the stops do not share a tribe.

In the conversations I keep having with people my age and younger, they keep saying the same thing: it is almost as if we are back in the Gen Z protests. Something else has entered those conversations, and I cannot remember the last time I heard it said of a Kenyan politician: I hope nothing happens to him. There are mothers praying that he lives to stand on a nomination platform and face the president. The prayers have their reasons; his rallies have been violently disrupted more than once. That is not the vocabulary of a fad. Each morning brings a fresh rumour of a campaign collapsing or a movement at war with itself, from blogs without bylines and papers with leanings, some toward the state, some toward rival opposition camps. Much of it cannot be verified; the pattern can. The man said little in public for months after Raila’s death and still led the coverage. He is the story of this season, and the attacks run his publicity free. Voices aligned to the administration keep reaching back to tribal politics against this wave, though it was their own camp that bent that model in 2022. The veterans of every count since 1992 reach back to it because it is the only model their data has ever confirmed.

Sifuna himself tells Kenyans it is not about him. The job, as he frames it, stays the same whoever carries it: make William Ruto a one-term president. Wantam, the crowds call it. Then his supporters answer him, in numbers, that if it is not him they will not wake up to vote at all. Some answer differently, that if he does not run the movement will simply find another young carrier. But the louder answer has been the stay-home one, and the movement’s whole risk lives inside it. Abstention is arithmetic for the incumbent. A wave that says it is about the idea, and then behaves as if it is about the man, is volunteering to prove the sceptics right. Sifuna is not the idea. He is the first vessel this idea has trusted. And the idea, for a season at least, is a hope with two strands. One is that tribe can be set aside for this one election, and that one election becomes a stepping stone to setting it aside for good. The other is that impunity can be ended once and for all. That strand runs straight back to June 2024. One of the defining features of the anger then was the opulence on display, a political class flaunting its wealth at the very people it was asking to pay more. Pride in ethnicity is nothing to apologise for. But ethnicity has been a scourge in this country at moments too, and that has always been its fine line: where does pride stop and become something else.

So look at the vessel. He was born in western Kenya in 1982, a Bukusu, the son of a teacher and a civil servant, and he is of my generation. He read law at Nairobi, was admitted in 2008, built his own practice, and stayed close to the Law Society. In 2017 he tried for Parliament at home in Kanduyi and lost before the general election even arrived. The home route, the one the old heads say a western politician must take, rejected him on its own doorstep. The party then threw him into the Nairobi Senate race that August, and he took more than six hundred and seventy thousand votes in the capital and still lost. Five years later he came back and won the seat, in the most tribally mixed county in the republic, with more than seven hundred thousand votes. His base turned out to be the cross-tribal city, and now a cross-tribal generation. Nairobi in both years was also Raila Odinga’s coalition in a city Raila carried decisively, so the live question was always whether that mandate was Sifuna’s or borrowed. The crowds the movement has drawn this year, with him at their centre and no party machine behind him, lean the answer his way. They do not settle it. Raila is gone, and some of what gathers may be his orphaned following looking for a home.

It is an old pattern: a man is often seen properly abroad before he is seen at home. Sifuna’s version has completed its circuit. In 2017 home and the capital both said no. The capital came round first, in 2022. This year home caught up: the declarations came out of the west one after another, and in the west, the sermons now find their way to him. The old lens will read those declarations as a bloc assembling. It should look at the crowds instead. Home has recognised him; that question is closed, and it is not what 2027 will decide.

What made him the vessel was a stripping. Raila Odinga died on 15 October 2025, in India, at eighty, still inside the broad-based arrangement he had signed with the man he lost to. The party he built stayed inside it. Sifuna would not. His refusal was categorical, and it is much of what endears him: he has said he will not support Ruto’s re-election under any circumstances, because the country cannot afford another five years of it. He threatened to leave the party rather than see it back the President in 2027. That stance is what the party punished. In February this year ODM’s national executive removed him as Secretary General, the post he had held since his mid-thirties; his own summary was that his only crime was opposing the plan to support Ruto. In July the Registrar of Political Parties approved the removal. The Senate deputy whip’s post went with it. He contested the process through the institutions, as a constitutionalist would, and he lost through the institutions, as men who challenge power usually do. The positions taken from him became his credential. His own answer carries the arc: some of those positions came from Raila, he said, but most of them came from the public. He is the truest heir of the Raila school and the loudest rebel against what Raila’s party became. While Raila lived, it was Raila’s hand that kept him inside it. Sad as the loss was and remains, this moment did not exist while he lived. That irony is difficult to reconcile, because I suspect Raila would have been proud.

On 26 September the Linda Mwananchi movement named him its leader and its presidential candidate. The vehicle is expected to be unveiled in these first days of October. Even the name had to be fought for. The Registrar refused to reserve Linda Mwananchi, allies accused the state of blocking the registration and took the refusal to the disputes tribunal in their own names. This week the tribunal ruled the refusal unlawful, and the Registrar’s reservation letter followed within a day. This time the institutions cut his way, and the manner of it matters: the movement answered the blow by filing, and it won at law. And the win came against a clock. The commission’s published election calendar requires parties to lodge the names of their authorised officials by 15 October, with polling day already fixed for Tuesday, 10 August 2027. The tribunal freed the name fifteen days before the first gate.

The veterans’ scepticism stands on a record. Until 1991 the question did not arise. Kenya was a one-party state by law from 1982. In 1988 the ruling party’s nominations, which decided the seats, ran by mlolongo: citizens queueing in public behind their candidate’s agent while officials counted heads. Loyalty was visible, and so was its price. The multiparty ballot returned in 1992 carrying a genuine wave, and the wave died by its own hand. The anti-Moi vote was a clear majority of the country. Then FORD broke into two parties along the very lines the movement claimed to have transcended. The majority split three ways, between Kenneth Matiba, Mwai Kibaki and Jaramogi Oginga Odinga, and Daniel arap Moi won with roughly a third of the vote. In 1997 the opposition arrived divided again, four ways this time, and Moi won again without a majority. The analysts drew the exact lesson the record taught: coalitions win Kenyan elections, and coalitions are assembled from ethnic blocs.

2002 is the exception they will cite, and it proves their rule as much as it breaks it. The NARC wave was real. Kenyans of every region voted against four decades of KANU, and Kibaki took roughly six votes in ten. But the wave rode a machine. Raila’s “Kibaki tosha” carried a memorandum of understanding among regional barons, and KANU had already imploded after Moi anointed Uhuru Kenyatta and his lieutenants walked out. The idea won in 2002 because the barons agreed to carry it.

2007 is the wound, and it is also the tell. Kibaki had presided over the strongest growth in a generation, corruption stories and all, and the economy still could not settle the ballot. The count collapsed into dispute, the swearing-in came at dusk, and the inquiry that followed ruled the true result unknowable. Many Kenyans ruled otherwise, and still do. The country burned along the same lines the parties had mobilised: more than a thousand dead, hundreds of thousands displaced, and in the end power shared under international mediation. The dispute’s arithmetic lived in a simple comparison, presidential votes cast against parliamentary votes cast in the same constituencies, numbers that should move together and did not. The same comparison, run on Zambia’s 2026 count in Polished Teeth, Blank Portal, found 186,247 more presidential than parliamentary ballots across twenty-one constituencies, every one won by the declared winner. The instrument traces back to Kenya’s 2007.

2013 followed as the census election: Uhuru Kenyatta and William Ruto, then facing ICC cases that later collapsed, fused their blocs and cleared the constitutional bar by a fraction of a point against Raila’s coalition. 2017 gave constitutionalism its high-water mark and its stress test in the same year. The Supreme Court annulled a presidential election, the first court in Africa to do so, and the re-run collapsed into a boycott. Through all of it, the blocs held.

Then 2022, the crack. Ruto won against the combined weight of a sitting president and the opposition’s eternal candidate, and he did it by running a class story against a coalition story. Hustlers against dynasties cut across enough of the old blocs to win, including deep inside territory his opponents assumed was theirs. The model bent in 2022; it did not break. And the man who bent it is the incumbent this wave now seeks to retire.

Kenya has run an idea before. In 1992 the idea was the ballot itself, the end of one-party rule. It commanded a majority, it arrived carried by rival barons, the carriers split it three ways, and the machinery won without one. In 2002 the idea was the end of KANU, and it won when the machinery defected to it: the barons unified behind one man and the ruling party imploded. An idea has never yet won in Kenya except by borrowing the machine. Even 2022’s class story ran on a machine of its own. What is different about 2027, on the evidence so far, is the carrier. This idea came up from a generation rather than down from barons, and the declarations have followed the crowds rather than summoned them. What is unchanged is the trap. The machinery has won whenever an idea’s carriers split, and it has won that way without ever beating the idea. The opposition today is still several camps in search of one ticket, and I am not convinced the leaders of the other factions have read the mood or understood the moment. The money question runs beside it. They keep telling him he cannot fund a national campaign, and on the arithmetic of agents and logistics the point is real; machines pay for elections. The reply forming is one Kenya has not seen at this scale: a ten-shilling paybill, and citizens, many of them poor, handing the movement their own money at its rallies. The sums are still small against a machine. And the standard is already biting its owner: a donor of fifty-five shillings has gone to court to compel an accounting. A movement running on the rule of law will be audited by it. The sceptics pricing a 1992 replay are not being cynical; they are reading the record, and they could be right. That is why the next twelve months fascinate: the record and the carrier have never been this evenly matched.

The constitution splits its verdict. A winning candidate needs more than half of all votes cast, and at least a quarter of the vote in more than half of the forty-seven counties. The first test sides with the record; majorities here have always been assembled from blocs. The second demands breadth, and breadth does not care how it is assembled. For thirty years coalitions of blocs were the only way to assemble it, so the test has been read as tribe. A genuinely tribeless candidacy is simply the second way through the same gate, and the easier one, if its crowds reach where the tour has not. And 1992 cannot replay exactly, because this constitution added a second round. If no one clears both tests, the top two meet again within thirty days. Kenya has never yet needed that round; the 2017 re-run was a court’s order, not the arithmetic’s. A fractured field in 2027 could force it for the first time, and a runoff is a different fight from the one Moi won with a third of the vote.

Then there is the roll, and it is the number that will settle the argument. The median Kenyan is about twenty years old; the country’s centre of gravity has never stood further from its ruling generation. The generation that breached Parliament registered poorly in 2022, and it began this cycle no better. By March the electoral commission was raising the alarm that two of every three new registrations were coming from voters over thirty-five. Then the mass drive in April closed with the commission declaring its two and a half million target met, the earlier slow months counted in. The late surge came from the young, and Nairobi led all counties. Registration has since returned to the commission’s offices, ticking along quietly. Which is why the weeks ahead are the first honest measure of this wave. A movement has named its candidate, and its vehicle is days from unveiling. If the wave is real, it will now walk to a registration desk, and that number cannot be spun. A wave that does not convert to the register is weather. Ruto is formidable, and the delivery is real. But if this generation registers first and then votes, the arithmetic that has organised every Kenyan election since 1992 stops organising this one. The crowds say which way it would break. The register will say whether it breaks at all.

And there is the man they must retire. He has spent three decades inside Kenyan politics, and he is the incumbent. I do not know that I have seen a Kenyan president work this hard or move this much. The development tours run three and four days at a stretch, sometimes longer: the South Rift, Luo Nyanza, Ukambani, the Coast. Dangote’s two trillion shilling refinery broke ground at Lamu the week he landed back from New York. Critics read the tours as early campaigning, and the expenditure, State House’s included, is a fair argument. That he does not work is not one. The economy he was handed in 2022 was close to the rocks; the one he governs today has been pulled well off them. The shilling that was sliding has held, exports have grown, and the external position has been repaired to a degree the 2022 numbers gave Kenya no right to expect. Imports have grown too, and the public debt has crossed thirteen trillion shillings, and on the law’s own present-value measure it stands in the mid-sixties per cent of GDP against an anchor of fifty-five. The external account has been well managed; it has not been cheap, and part of the repair ran under a Fund programme that lapsed last year with its final review unfinished. The economic team around David Ndii has been, to my eye, among the best on this continent, and it draws criticism even for the things it has plainly delivered. The delivery runs past the ledgers too: fertiliser at a third of its 2022 price, more than a hundred thousand housing units under construction, tens of thousands of teachers hired. Some of what Ruto opens was conceived before his time; completion still counts. Some of it speaks directly to the generation in the streets: a continental football championship in 2027, the world’s athletics in 2029, Kasarani mid-rebuild for the first and now needed for the second. A new national stadium is to carry Raila Odinga’s name. How some of this is financed is a separate argument, made in Converting Vices into a Virtuous Sporting Cycle. The building is real, and the credit is due.

All of it is true, and it may still not be enough, because the wave has two arguments and the aggregates answer neither. The first is diffusion. The headline has healed faster than the household, and the young are the ones formal work has not yet absorbed. The country’s own survey counted more than eight hundred thousand new jobs last year; nine in ten were informal. Quick Comment on Kenya’s Economic Balancing Act set Kenya’s external stability against its internal strain in January 2025 and named the 2027 election as the flashpoint if the gap went unclosed. Growth Without Diffusion found the same divergence in Zambia fourteen months later: growth returns to the ledgers before it returns to the street. Nor is it an African story. Diffusion failure is the argument of this age everywhere. It is why nationalism is rising through the old industrial democracies and why socialism is back on ballots that had buried it. Electorates now vote on the distance between the ledger and their lives, and Kenya in 2027 will be one more instance. The second argument is the constitution, and no delivery answers it. Ruto can win the argument about aggregates and still lose on diffusion and on obedience in the same afternoon. The ballot question has moved from who delivers to who obeys, and who feels it.

The same man who calls the wave a fad gave me one of the most interesting readings of Ruto I have heard. Above everything, he said, the man is a doer. He starts from whether a thing can be done and how to do it, and only then asks about the rules. Coming from inside the President’s own community, and as an explanation of the delivery record, it works. It also explains the collisions with the judiciary, and the moments when constitutional process has been treated as an obstacle standing between a decision and its execution. A constitution is built to hold a government back; it must be followed anyway. But when small things are skirted, the question becomes what happens with the larger ones. The wave’s case sits inside that reading, and it came from the same sceptic who calls Sifuna a fad.

Listen to the commentary long enough and you will hear that Sifuna has no agenda. It is the strangest charge of the cycle, because the agenda is the largest one on offer. A return to the rule of law and a state bound by its own constitution, where speech does not cost the speaker his liberty or his life. Kenya has voted on development, on unity, and once, in 2022, on class. It has voted on writing constitutions too: the referendums of 2005 and 2010, and the reform promises that rode the 2002 and 2007 campaigns. What has never led a presidential ballot is the narrower, harder question of whether the constitution already written will be obeyed. That is the entire offer here.

Some say June 2024 is too far back to decide anything in 2027, and they would be right if June 2024 had ended. It has not ended. The record since has kept teaching the same lesson, and the generation that recited the constitution’s articles in the streets knows exactly why the offer matters now. Albert Ojwang, a teacher who posted about a senior policeman, was driven three hundred and fifty kilometres to a Nairobi cell and did not leave it alive. The police said he had hurt himself, and the autopsy said otherwise. The June 2025 anniversary marches left nineteen dead by the human rights commission’s count; Saba Saba, twelve days later, at least thirty-eight more. The disappearances of critics since 2024 sit documented in the same commission’s files.

Then there is the quieter thing that may cut deepest, because it touches the one economy this generation built for itself. Locked out of formal work, the young moved their labour online: content, code and commerce run from a phone. It is thin ground. The Loud Ten Per Cent found six in ten African creators earning under a hundred dollars a month, because the platforms price an African audience at a fraction of a Western one. The state’s most visible arrival in that economy has been a deduction. Under a tax written in 2023, the platforms now hold back five per cent of whatever they pay a Kenyan creator and pass it to the state. Meta began the deductions; Google followed, and its deadline for creators to hand over their tax PINs fell the day before this essay publishes. The deduction moved faster than the support: the bill meant to grow this same economy trailed years behind the tax that draws from it. The young built a room the state never furnished, and the first knock on the door was the revenue authority’s. The taxes are lawful and the order is necessary. That account is still the one a generation carries to the register.

Zambia taught me one more cold thing this year. The world will not referee 2027. The Forced Choice set the rule out in February: the referee era is over, and the countries in between choose deliberately or are chosen for by default. Nairobi and Washington are warm again, and great powers go quiet where their interests sit comfortably; the hardening of the continent’s political space, which After the Count traced from Lusaka, grants Kenya no exemption. If the count is disputed and the response turns hard, nobody is coming. Zambia’s count carried a narrower lesson inside that one: when the powers that be sit with the incumbent, a close win is not a win. An opposition in that position must take the race beyond dispute. What Kenyans hold that Zambians did not is their own architecture. A stronger constitution, a court that has already annulled a presidential election, and a public that has protested for what is right and died for it. And history built one deterrent of its own. The last count disputed to the point of fire ended at The Hague, and nobody on any side wants the sequel.

The old lens hears all this and asks where the machine is, and the question has a precedent it should sit with. In 2008 the United States elected a forty-seven-year-old with four years in the Senate and a thin file by the standards of the men he beat, because he carried an idea into its moment. The idea built him a machine of its own, small money and volunteers, inside an established party. Sifuna is building his outside every established party, from a ten-shilling paybill up. In an ordinary year, the machine wins. An idea is not undefeatable; 1992 buried one on its own doorstep. But an idea that arrives with its moment is the hardest thing a ballot has ever had to stop. By the old lens, 2027 will be an ordinary year. This time, I am not sure the old lens is right.

One more photograph, then. In late 2014, as Zambia’s ruling party fought over a dead president’s succession, a single frame captured the country: one contender sat on the grass under a tree while his rivals sat on chairs. Zambians named him the humble man, and two months later Edgar Lungu was president. Kenya carries an even more powerful image, and it comes from 25 June 2024, one of the most extraordinary days in living memory. I suspect that day will end up a public holiday, depending on who takes the presidency. A senator running at armed police to stop them shooting at the young. Expect it to be played and replayed from now until the ballot. And it will not replay alone. On that day Kenyans watched, live on television and on their phones, young people shot dead at the gates of their own Parliament; in the seasons since, the dead have included children. A broadcaster later reconstructed the Parliament shootings frame by frame and identified one shooter as a serving soldier; the authorities’ answer was to block the film’s screening. The frames survived. They are archived on millions of devices, and they will replay too. The President’s side has pictures of its own, and they are real: the roads, the stadiums, the cranes on the skyline. That is the choice, told in images: what a government builds, against what a government obeys. Building and the constitution have led Kenyan ballots before. The rule of law never has. If 2027 has a defining frame, it is already filmed.

AI-illustration: The Two-Sided Coin

Inflation expectations in Zambia are broadly unanchored and adaptive, and the exchange rate and the policy rate are their key drivers. Restated from Bank of Zambia Working Paper WP/2026/2.

1. The Week the Money Left

The record is six rows of a table that took 66 days to appear (Table 1).

The Bond B buyback, the retirement of the 2053 restructuring bond, settles in these rows. The USD599.82m arriving in the week ending 12 June is the African Development Bank facility. The USD1.12bn leaving in the following week carries the tender payment. The instalment on the restructured 2033 bonds follows at month end. Net of the facility, the operation drew USD521.76m from reserves on these rows; the Bank’s Q3 presentation, released with the rate decision, states the cost as USD514.8m and reports July reserves recovering to USD6.0bn.

The tender’s terms were public when “The Most Expensive Recovery” (Canary Compass, June 2026) ran them through: 84 cents all-in, consideration, fees and accrued, on the USD1.365bn outstanding, against the USD600m AfDB facility. The result those terms implied, roughly USD546.6m of reserves at full uptake, was the projection, published ahead of the results of 10 and 12 June. Table 2 grades it against them and the settlement rows.

The buyback variance decomposes cleanly. The AfDB leg was exact to within USD180k. The payment row unpacks first: at the notes’ 0.5 per cent coupon, consideration, fees and accrued rebuild to about USD1,109m, leaving roughly USD12m of the week’s routine service inside the row. The tender record settles the rest: 97.85 per cent accepted at the early deadline, at USD828.68 per 1,000 plus accrued, about 83 cents all-in. The miss against the June arithmetic’s 84 is uptake with a cent of price. The clean-up call retired the remaining 2.1 per cent by 25 June, at the tender consideration plus accrued, the early fee excluded.

The recovery also grades the design. A net draw near USD515m was refilled within a month, and the market barely felt it: the interbank mid averaged 17.83 in the payment month and 18.34 in July, both stronger than May’s 18.85. The interbank is the right ruler for reserves: the Bank deals there, while the slate and retail series elsewhere in this note price the pump path. July reserves came back to USD6.0bn, 4.5 months of cover, on mining taxes and project inflows; the reading is the Bank’s September presentation, ahead of the fortnightly print. The engine is the one “The Most Expensive Recovery” rested its case on. The second quarter’s current account surplus widened to USD0.3bn from USD0.1bn, and the mining sector supplied USD1.2bn of foreign exchange in the quarter. The offshore bond bid had added USD674m of market supply across the first half, before its primary room ran out.

On 7 July, about three weeks after settlement, Governor Denny Kalyalya told the public not to downplay Zambia’s reserves, citing USD6.5bn and five months of import cover. The Bank’s own later presentation puts end-May cover at 4.9. That was the end-May position, the latest the Bank had published; the June drawdown was settled but not yet in print. The Bank’s own 30 June position, on its own later presentation, was USD5.80bn at 4.4 months of cover, and the publication carrying it is dated 4 September. The statement was accurate on the public record and five weeks older than it sounded. The fuller context was the Bank’s to give, and the fresher number crossed an election on its way to print.

2. The Books Nobody Restated

2.1 The Publication Record

The Ministry of Finance publishes its fiscal outturns in the Monthly Economic Indicators. The 2026 series is not in good order (Table 3).

Four of the seven fiscal editions print corrupted tables. The framework is the newer failure: Parliament replaced the budget in May, and every edition issued since still scores against the one it replaced. The February truncation was on record in “Seven Stars That Refuse to Align” (May 2026); the rest accumulated after it. The headline cell a reader would lift from the most recent one is wrong as printed: the fiscal balance reads as a surplus under its own Surplus(+)/Deficit(−) header. Beside it sits a variance of 693.5 per cent, computed on the sign-dropped absolutes. The July fiscal balance is a deficit, quoted here with its subtraction shown: 114,281,456 minus 138,443,049 equals minus 24,161,593, in thousands of kwacha, confirmed by the financing table beneath it.

2.2 The Reconstructed Record

Table 4 rebuilds the monthly ledger from the archived prelims.

Three readings follow from the table. First, July, the month before the 13 August vote, was financed ZMW2.0bn from securities and ZMW3.9bn by draining government deposits; smaller lines net the difference. The buyback does not explain the drain: both its kwacha legs and the payment out are booked in June’s monthly financing column and net there; July’s buyback lines print zero. The buffer built from January to May was spent going into the vote. Second, June’s near-zero fiscal balance has two supports, neither of them spending discipline. June is the company tax provisional month, whose profile targeted a fiscal surplus. The month also booked ZMW2.02bn of unbudgeted revenue under the label Exceptional Revenue, a non-tax line in the MEI’s Table 1.0 with no budget provision and no explanation, now ZMW2.41bn for the year to July. Strip it and June’s fiscal deficit is roughly ZMW2.9bn; the strip treats the line as non-recurring until the Ministry classifies it. Third, across June and July the Ministry’s own profile budgeted a net fiscal surplus near ZMW0.95bn; the outturn was a ZMW6.6bn fiscal deficit. That is a ZMW7.5bn deterioration against plan in the two months before the vote. On the ceilings the distinction is pace against breach: the original was exceeded outright in July; the revised is being consumed at 129 per cent of pace, with ZMW7.99bn of room for five months.

The primary balance is the programme variable, and its path tells the year in four numbers. It ran 3.1 per cent of GDP in 2025 on the Fund’s series. The 2026 budget targeted 4.2 per cent on its own GDP base; on the statistical bases that is 4.5 and 4.3 per cent. The IMF’s May projection sits at 1.1 per cent, cut from its own prior 3.8, naming the fuel suspension, election spending, the wage settlement and the agricultural overrun as the drivers. At July the outturn is +0.76 per cent on the published quarters, +0.72 on the full-year estimate. Seven months in, the outturn tracks the Fund’s revision, not the budget’s ambition.

2.3 The Budget Zambia Actually Has

Parliament approved a ZMW26.3bn supplementary on 11 May, and the Supplementary Appropriation Act, assented on 4 June, prints the allocation: 84 programme amounts across the heads, summing to ZMW26.32bn. The MEIs ignore both. The schedule’s two largest finance lines are a ZMW2.90bn Centralised Holding Vote and ZMW2.68bn of Centralised Strategic Payments. A vote is the budget’s holding account, a line that carries money without naming a programme; these two carry ZMW5.6bn between them, unnamed.

Three outside transcriptions exist, and ours reconciles to all of them. ZIPAR, the Zambia Institute for Policy Analysis and Research, with UNICEF, transcribed the Estimates at head level in its mid-year analysis; CSPR, the Civil Society for Poverty Reduction, adds a sector summary. We combined both with the reconstruction, and the Act corroborates it: the Social Cash Transfer and Food Security Pack increments match exactly at ZIPAR’s printed precision, the agriculture head within ZMW5m. ZIPAR’s half-year revenue figure of ZMW96.3bn reproduces our monthly sums plus the ZMW1.73bn of unannounced revisions we had localised by differencing consecutive editions. ZIPAR’s July cash transfer figure of 56.87 per cent of the revised allocation equals ours to the second decimal. The ledger is internally sound; the publications describing it are not.

What the combination cannot do is make the supplementary’s own arithmetic close (Table 5).

The residual moves with the reading; on neither published basis does it close. Add up every source of money the government has named for this supplementary, and it does not reach the amount Parliament approved. ZIPAR’s transcription of the Estimates financing table prints the borrowing increment as ZMW10.36bn rather than the ZMW7.5bn of the ministerial record. On the reading that the larger figure subsumes the ZMW2.4bn of carryovers, the residual is ZMW2.03bn; counted separately, the schedule would over-finance, which no account claims, so the classification is unresolved. We publish on ZMW7.5bn, the basis of record, and carry the variant as a question about a document never published in full. On that basis ZMW2.5bn of the appropriation has no named source; the carryover reading narrows it to ZMW2.0bn. The ministerial presentation separately counted a delayed disbursement under the IMF’s Extended Credit Facility (ECF), USD95m received after slipping from late 2025, about ZMW1.8bn at the May interbank. No transcription’s financing table carries it; counted in, the residual narrows toward ZMW0.7bn, and the reported domestic-financing figure itself varies between 5.75 and the 7.5 carried here. No reading closes the gap, and no single published table reconciles the financing.

On the spending side, CSPR’s analysis of the Estimates carried ZMW7.1bn, 27 per cent, as unspecified. The enacted schedule has since named every programme; what stays unconfirmed is the mapping: the announced purposes reconcile to the two centralised votes arithmetically, in Table 5’s note, and no appropriation document maps them. The ZMW10.05bn of declared savings is invisible twice: no document, the Act included, names the heads that surrendered it, nor how much each gave up, so every revised allocation is an upper bound. And the widely reported ZMW279.4bn revised total, the ZMW253.1bn original plus the ZMW26.3bn gross supplementary, ignores the offset. The net envelope is ZMW269.4bn. Nearly five months on, its financing does not add up in public, the two centralised votes still name no purpose, and the Ministry still scores the year against the budget it replaced.

2.4 Execution Against the Revised Budget

Table 6 screens the July outturn against both the original and the revised allocations.

The restatement cuts both ways. The elections line, scored exhausted against the original budget before the vote, was funded through it once the supplementary doubled the ECZ head; the revised basis is the right one and Table 6 adopts it. The original line was written for a smaller election: the supplementary’s ZMW1.14bn increment funded the constitutional amendment’s new map, 226 constituencies from 156 and 16,400 polling stations from 12,152, on CSPR’s transcription. The maize line survives every restatement: nearly nine tenths of an envelope raised by two thirds, consumed before the input season. And the external service line shows the buyback for what the fiscal accounts say it was. The ZMW20.04bn that moved through June’s books nets to zero across the financing table: the ZMW10.4bn AfDB loan and the ZMW9.6bn Bank of Zambia leg in, the payment out. The kwacha legs are section 1’s dollar legs: the loan is Table 1’s USD599.82m facility; the Bank’s leg matches the official USD514.8m drawdown. The loan books at 17.40 kwacha per dollar, the market of its own drawdown week: 11 and 12 June printed the month’s lows. The Bank’s leg books near 18.65, above anything June traded. On the Bank’s published dailies, June ran 17.37 to 18.33 around a 17.83 average; 18.65 is a May-range rate. The operation financed none of the fiscal deficit. Strip it and external service falls to 44.9 per cent of the revised line, behind pace rather than through it.

2.5 Where the Money Went

Total expenditure excluding amortisation printed at 99.7 per cent of its prorated pace at July. The aggregate conceals the line-level divergence (Table 7).

The composition is the table’s: political lines exploded while roads, the CDF, school feeding and pensions were starved, and consumption taxes were administered away. Mining collections, past their full-year line by July, bridged what the rest gave up. That bridge rests on one commodity. A copper economy at record prices is simultaneously the budget’s largest income tax payer and a structural source of its VAT refunds, because exports are zero-rated and copper is the export base. The appreciation that flatters the external accounts compresses every kwacha the state converts, the trade-off set out in “Copper Output and the 2026 Royalty Arithmetic” (January 2026).

Three unnamed items remain. Exceptional Revenue, ZMW2.41bn with no budget line and no name, is 2.1 per cent of revenue to July. A recapitalisation line is 31 per cent over its full-year allocation with the recipient unidentified. And a ZMW458m upward revision to grants has appeared in three consecutive publications, attributed to no month, discoverable only by differencing editions.

2.6 The Maize Position

The Strategic Food Reserve line is the centrepiece of the screen, and its mechanism sits outside the budget (Table 8).

The 2025 gap was funded off the books: roughly ZMW5bn of commercial borrowing collateralised by the maize itself, a ZMW2bn Zanaco facility, and ZMW3.3bn of unfunded arrears. The Secretary to the Treasury’s April presentation acknowledged the arrears. The supplementary’s ZMW7.4bn agriculture allocation, ZMW5.0bn to food reserves and ZMW2.4bn to the input programme on the Act’s schedule, was not new spending; it cleaned up last season’s obligations. The MEI’s Strategic Food Reserve line, ZMW7.2bn by June and ZMW9.9bn by July, is that cleanup plus the new season starting. The new season then outran every frame, on the agency’s own account: a record 2.03m tonnes in under two months, 2.73m projected. On the pledge to pay within three days of delivery, roughly ZMW4.9bn falls due within days as deliveries are accepted, against appropriations already close to spent. The overflow heads for bank balance sheets, secured on grain, the same balance sheets the auction book clears through. The ZMW25bn that circulated in the market as the agency’s funding appetite was never published with a composition. Published tonnage and gazetted prices put the new season alone at roughly ZMW19bn and the two seasons at ZMW30.3bn, and the rumour sits inside that bracket. The agency chair’s own season figure, a record ZMW20bn as marketing closed, counts maize and 100,000 tonnes of rice together; the maize-only ZMW19bn sits inside it. Counted with the rice, the two seasons head past ZMW31bn.

The collateral does not cover the cost. Procurement cost runs near USD355 to 376 per tonne across the two seasons at the year’s exchange rates. ZAMACE market reporting, the Zambian commodity exchange, put the agency’s offer price near USD275 and the regional market near USD230 in September. The one large export deal, 540,000 tonnes signed with Kenya’s Baita Trading on 9 September, delivers in six tranches at undisclosed prices. Even at USD230 to 275, the full deal recovers ZMW2.4 to 2.9bn against a two-season procurement bill heading past ZMW30bn: under a tenth of the bill. Every tranche priced below cost would realise part of that loss; the undisclosed prices withhold the margin, not the exposure. No public record states whether the Zanaco facility has been repaid. The agency’s board put the old season’s unsold stock just over 1.4m tonnes in September; with the new season’s 2.73m, the holding heads past 4m tonnes, bought above its selling price.

Storage is where the arithmetic stops reconciling. The agency’s own secure capacity is roughly 1.5m tonnes, per “Seven Stars That Refuse to Align” in May; the Agriculture Minister put national capacity, 375 facilities including 72 private, at 1.83m the same month. The executive director’s mid-September count corroborates it: 3.1m tonnes held on 18 September, 1.7m new and 1.4m old, already more than a million beyond the national figure. No published record says where the balance sits: a property director sent to optimise storage space, satellite depots nationwide, and last season’s open-air precedent are the agency’s own markers. That precedent ran near a tenth of today’s overhang: roughly 170,000 tonnes outside secure capacity then, 1.67m bought against 1.5m of room, against 2.3m beyond the national figure now.

The position is also provisioning, and the price risk it insures runs one way. The strategic reserve target is 2.5m tonnes, the Agriculture Minister’s January figure. The stock stands ahead of an El Niño season NOAA’s Climate Prediction Center carried at better than 90 per cent odds in August, as “After the Count” recorded. If the drought lands, maize prices move up, not down. A failed harvest would meet a reserve drawn into domestic need and exports already contracted, and the mealie-meal price the CPI reads could run far past today’s floor. In that market the holding is the brake, and the below-cost arithmetic above inverts: scarcity reprices the stock toward what it cost. The purchase also has a beneficiary already banked: farmers harvested into a guaranteed buyer at a gazetted floor above what the grain resells for, the political win “Seven Stars” recorded in May.

Which reading carries is measurable: the pace of releases and export tranches against last season’s, where 1.4m of 1.67m was still held a year later. The ZMW30.3bn is a two-season procurement flow, not an outstanding debt; the bank exposure, unquantified while old-loan repayment stays undisclosed, builds in the new season’s unfunded share. Positions like this surface in the fiscal tables eventually: as arrears, as a recapitalisation, or as a supplementary. It is the fuel stabilisation pattern applied to grain, a discretionary quasi-fiscal operation visible in the accounts only as the settlement line.

3. The Financing Triumph

Positions that size end at the same constraint the Treasury itself faces: the financing requirement. In July we circulated a private note, “The Year as a Single Equation”, which reduced the year to one requirement and ended with four doors. This section restates its equation on the year-end debt bulletin’s 2026 schedule. Maturities at cost of ZMW85.86bn, from Table 27 of the year-end Debt Statistical Bulletin, plus the ZMW21.62bn issuance plan give a requirement of ZMW107.48bn original and ZMW114.98bn revised. First-half delivery was ZMW58.44bn. The note’s own working figures, and the ZMW106.07bn carried earlier from the Annual Borrowing Plan, predate the year-end bulletin; the figures here supersede both. Crediting the ZMW4bn opening balance in the proceeds account, the second half had to place 83.7 per cent of every kwacha offered on the original framework, 97.7 per cent on the revised. That is the placement test; the delivery shares below are a different ratio. The balance is not a published figure; it reached us through the private reconciliation the note was circulated for, and it is material: without the credit the tests read 91.1 and 105.1 per cent. The offer base is ZMW53.8bn: Q3’s announced ZMW25.8bn plus the ZMW28bn assumed below for Q4. When the note circulated in July, trailing four-month delivery stood at 79 per cent of offers; the same window ending September stands at 110.

The four doors, with no fifth. Issuance delivers more, through bigger offers, fuller take-up or allotment beyond offer; maturities are bridged from revenue; the net domestic financing (NDF) target is missed; or the debt stock shrinks.

The third quarter answered through the first door, by allotment beyond offer rather than bigger offers: the calendar stayed scarce and the window took 117.7 per cent of what it showed (Tables 9 and 10).

What remains, net of the proceeds-account credit, is ZMW14.68bn of the original requirement and ZMW22.18bn of the revised. Against an assumed ZMW28bn of fourth-quarter offers, that is 52.4 and 79.2 per cent; without the credit, 66.7 and 93.5.

Across the three auctions roughly ZMW19.2bn of face value chased ZMW10.7bn of offers; ZMW11.4bn was allotted at cost, over offer, and ZMW7.9bn of face was turned away. Since January the 91-day is down 300 basis points, the one-year 360, the bond curve 269 to 334. The mechanism matters more than the levels: the requirement collapsed first, the rejections became cheap, and the cut-offs followed. In March and April the Treasury rejected bids with the requirement still unmet, compression taken on credit, and April’s bond auction cleared at 20 per cent of offer to show what that costs. By September the requirement was largely closed and the same rejection tool was operating on cash. Rejecting price while banking quantity is market power, and it is why these lows are sustainable into the fourth quarter in a way March’s never were.

The latitude is arithmetic. Delivery stands at ZMW88.8bn, 83 and 77 per cent of the original and revised requirements, with a quarter to run. The Treasury has latitude to refuse expensive bids and wait, within its cash position and the maturity calendar. The NDF ceiling pushes the same way: at 95 per cent consumed, net issuance has almost no legal room, and rejection costs nothing against the financing provision. Both forces run yields down while the fiscal record argues the other way.

The cash came partly from the budget’s own front-loading and partly from an administrative act. On 31 July, between MPC meetings and thirteen days before the poll, the Bank cut the statutory reserve ratio (SRR) on kwacha deposits from 26 to 21 per cent by circular. Foreign currency stayed at 26, and compliance moved from a daily test to a weekly average. The Bank’s liquidity fortnightly dates the release: statutory reserves fell ZMW6.5bn in four trading days, parked first in term deposits, then surfaced at the auctions. The bill sale three days after the release drew 2.18 times its offer; the one a fortnight later, competing with a bond the next day, drew 0.78.

The instrument has a published design, and the August version was half of it. “Zambia Monetary Policy, Part 3” (Canary Compass, December 2025) proposed that when this cut came, it should come with the relief earned rather than granted. That condition was the point, if the target was truly to expand credit access and boost private sector growth. The design had two arms. Relief against lending: 200 basis points at a 50 per cent private loan-to-deposit ratio, 400 at 60. And an asymmetric ratio that raised the foreign currency requirement while settling it in kwacha, so the release de-dollarises as it eases. Either arm is built to stop freed liquidity simply rotating back into government paper. August delivered the relief with both arms deleted: five points of kwacha relief, no lending condition, no foreign currency leg. Unconditioned, the release can still buy growth, but public-led growth, because in this system freed liquidity has a default destination, and the fortnight’s data showed it arriving there.

The dashboard the design was built on says why targeted beats blanket. Kwacha private credit has slipped to 7.0 per cent of GDP at July, from 7.4 published a year earlier, 7.2 restated on the final GDP; total private credit is down from 13.6 to 12.0. Credit to the private sector, the thing a cut is meant to buy, was shrinking relative to the economy. The other side of bank balance sheets says where the capacity sits. Government securities stood at 44.5 per cent of ZMW132.1bn of kwacha deposits on the 18 September fortnight, lower than last September’s 51 print, with deposits up 26 per cent in the year doing the work. The ratio is still elevated, and the comparison flatters the fall: last September’s 51 was read after that month’s bond auction had settled, while the 18 September fortnight sits before this one’s. The roughly ZMW6.3bn the late September auction raised at cost settled as the month closed, taking the ratio back toward 49. Kwacha private credit sits at 47.1 per cent of deposits, below the 50 per cent floor Part 3 set for earned relief. That is the case for the targeted instrument: a reserve release conditioned on lending attacks the 47.1; an unconditioned rate cut feeds the 44.5, the 49 on the settled reading. The October monetary statistics grade the split: private kwacha credit against bank holdings of government securities.

The tenor split points to who is buying the curve at these levels. At the 25 September bond auction the two-to-seven-year bucket took ZMW6.35bn of bids at cost against ZMW3.08bn at ten and fifteen years. Cost runs above the face bids in Table 10 because the outstanding coupons price above par at these cut-offs; the reopened seven-year allotted at 117.5 per 100 of face. Against August, the front grew 36 per cent and the long end shrank 38. The local bid replaced the offshore one at the margin, with the primary room spent and the withholding arithmetic already favouring the local book. A non-resident faces 20 per cent WHT, so the one-year at 10.90 gross is 8.72 net. Against the full-tax counterfactual for August, the index with the suspended fuel taxes restored, near 7.6 (section 4 derives the series), that is a real return of about a point. A local bank reclaiming the tax clears about three points on the same index before corporate income tax, and more than four on the printed one. At these cut-offs the arithmetic favours the local, and the Bank’s Q3 presentation says both halves of this in its own words.

Foreign institutions’ foreign exchange supply fell from USD542.3m to USD132.1m in the second quarter because, in the Bank’s phrase, their allocation in the primary market was exhausted. Their holdings tell the other half, at the June vintage the Bank has published: ZMW78.2bn, 28.9 per cent of the ZMW270.7bn outstanding, down just ZMW0.1bn in the month. The money that came in stayed; what stopped, on the data to June, was adding at the window. Whether they have added since through the uncapped secondary, or through local banks buying and on-selling, the grey area the cap does not police, is not observable until the next holdings print. Bond auctions were consolidated to two per quarter from monthly at an unchanged quarterly volume, so each window now meets more of the quarter’s demand at once. Three forces produced the decline: rejections, a reserve release, and a front-run rate cut. Each was administered or one-shot.

Two constraints sit under the triumph. Netted for the buyback pass-through, deficit-financing domestic borrowing stood at ZMW27.4bn by end-July: 128 per cent of the original ZMW21.4bn provision and 95 per cent of the revised ZMW28.9bn. The distinction needs stating precisely: gross issuance against maturities is not capped, and the auction record shows no funding constraint. What is nearly exhausted is the net deficit-financing provision in the budget law, the borrowing Parliament has authorised to be booked against the deficit rather than against rollover. That constraint binds in the National Assembly, not at the auction window. What remains of the provision is ZMW1.5bn against ZMW7.99bn of deficit room. Any deficit beyond it needs financing from outside the line: deposits, external inflows or revenue overperformance, channels open but not the year’s method. The year’s method has been to borrow through the original provision and reject the expensive bids; widening the provision again takes a second supplementary.

And December is a wall on both sides. The external side is the restructured bonds’ December payment: they pay each June and December. June’s regular instalment was ZMW5.3bn on the July MEI, the ZMW25.3bn of external amortisation less the buyback’s ZMW20.04bn pass-through, and December carries the mirror. Its kwacha cost at the end-September interbank, 19.55, runs about 10 per cent above June’s 17.83 average. The domestic side is the December maturity tower. June and December stood as the opening profile’s two towers, as the January absorption essay mapped; June’s came in at ZMW9.1bn of face value, the year’s heaviest month. December’s count is closed on the auction record: ZMW9.0bn is dated into the month, last December’s 364-day bills, the December-dated bond lines and ZMW3.4bn of the year’s own issuance. Nothing on the remaining calendar adds to it. The 1 October auction’s own 91-day, settled the day after the auction, matures on 1 January, and the quarter’s bond auctions issue at two years and longer. December lands a tenth of a billion short, and June keeps the year’s heaviest month. What December carries instead is composition: ZMW1.0bn of June’s print was bill recycling, and December holds more than three times that. Nearly three in ten kwacha of the stock sit with non-residents whose primary allocation is spent. The cap is a flow limit, 23 per cent of the ZMW50.4bn gross primary bond issuance: ZMW11.6bn for 2026 against ZMW10.5bn of non-resident maturities, sized to roll the principal, as “The Most Expensive Recovery” mapped. The other ZMW14.1bn owed them, coupons and discount, is serviced from revenue, not rolled.

What the cap did not anticipate was the appetite: holdings rose ZMW12.5bn in the first quarter alone, across primary and secondary, and the room sized to roll the maturities was spent on adding by mid-year. The principal still comes due. Rolling it now runs through the secondary market, on a domestic balance sheet willing to intermediate, at a price the cap does not set. From 2027 the plan lowers the primary room to 15 per cent.

The fourth-quarter calendar, published as this note closed, answers: two bond auctions of ZMW6.3bn each, on 20 November and 18 December, with seven bill auctions at ZMW2.2bn, ZMW28.0bn of offers in all. That is the maintained reading, and it is the ZMW28bn this section assumed. It also adds a tenor. The November auction debuts a 20-year bond, the first new point on the kwacha curve since the 15-year arrived in 2007. “The Most Expensive Recovery” extrapolated a hypothetical 20-year in June on the curve’s 20-basis-point slope. Its decision boundary, on a gross basis against the 2053 at its June price of 84 cents, the bond the buyback then retired: roughly 9.4 per cent annual depreciation. A lower debut yield lowers that boundary with it. September’s curve runs the same arithmetic: 14.50 at ten years, 15.45 at fifteen, 19 basis points a year of slope, a debut at 16.4 if the slope holds and above the 15-year by construction. At these yields the tenor adds term, not duration: a 20-year par bond near 16 carries a modified duration close to six years, barely above the 15-year’s, because the coupon returns the money early. September’s buyers sat short: 44.9 per cent of bond sales were three years or less, and the average sale 7.2 years. Short sales mature in 2028 and 2029; a 20-year gives that wall somewhere to go, and allocations (an opening and a reopening) across the quarter's two dates would build the new line toward index-eligible size.

The offshore buyer, if it returns, comes through that secondary market, which now carries an index bid: the kwacha’s place in JP Morgan’s new frontier local-currency index was reported in late September.

The continent’s long ends say why the bid may come. In September, Kenya’s 20-year line, twelve and a half years left on it, cleared at 13.61, its new 30-year at 14.24, against 6.8 inflation. Uganda’s 20-year, new this year, was held at 15.00, with over a trillion shillings of bids rejected, against 4.6. South Africa’s 20-year marked near 9.4 at end-September, against July’s 4.3. Nigeria’s long end paid 16.8 in September against August’s 15.4 inflation, its one-year money 16 to 17: nominal yield without real yield.

Zambia’s 15-year at 15.45 against 6.1 is a real yield above nine on printed inflation, and a 20-year at 16.4 would be the deepest market-cleared real yield on the long curves priced here. On the note’s own 7.5 counterfactual it still runs 8.9, and the seat holds. The one deeper print, Uganda’s, was administered by rejection, allotment rationed to defend a price, where September’s Zambian auctions allotted beyond offer at theirs. What prices the risk is the kwacha itself. The Bank’s real effective exchange rate index rises on depreciation, so lower readings are stronger. May and June sit about a third below the long-run mean on the fortnightly’s own series, the strongest prints it holds; section 4 carries the reading. An average is not an equilibrium, and the index can revert through a weaker kwacha or through inflation running above partners’. The breakevens here price the first path; the second erodes the return instead. The full-reversion case is run because it is the hardest test, not the expected path.

The arithmetic of erosion runs at the extrapolated debut, and Table 10A carries it. The live dollar comparator is the 2033, the restructuring bond still outstanding, quoted near 7.5 mid on 1 October, four tenths above its level on announcement day. The tenors do not match, six and three quarters of remaining life against twenty; the horizons nearly do, the kwacha coupon compressing the 20-year’s duration to about six years. A dollar yield moves every bar: each point available elsewhere lowers the annual breakeven by about a point.

The house corridor, 22 to 24 in the medium term, annualises near 3 per cent: inside every bar. Full reversion of the Bank’s real effective exchange rate index, June’s 75.3 to the 117.7 long-run mean, is a 56 per cent move, before inflation differentials. Taken by the duration horizon it runs 7.7 per cent a year, a kwacha near 31: through the withheld 5.2 and the treaty 6.6, six tenths inside the gross 8.3. Spread to maturity it is 2.3 per cent a year: full mean reversion never catches the bond held to term. The index prices the drawdown, not the destination, and the instrument reads as built for the holder who can wait: the pension balance sheet, the tax-reclaiming local book. The window cannot prove the appetite, and the on-selling grey area blurs who shows up. The 20 November debut prices five days before the Committee meets; the December note grades the clearing yield, the offshore share and the tenor mix, and re-runs these boundaries at the debut’s prints.

4. The Price the Public Sees

The Committee sat on 28 and 29 September against inflation at the floor of the Bank’s 6 to 8 per cent band: 6.2 in August, 6.1 in September. The number is real. Its construction is fiscal, in three layers.

The first layer is the tax suspension: excise at zero and VAT zero-rated on petrol and diesel since April, Treasury instruments under the tax statutes, expiring 30 September. The second is the discretionary stabilisation line inside the Energy Regulation Board’s (ERB) wholesale build-up, an amount set each month at the Board’s discretion with no published rule. In September it posted its largest payout on record and pinned all three products’ wholesale prices at August’s level to the ngwee; pumps printed 25.29 for petrol and 26.86 for diesel through August and September. The third is the exchange rate slate, set by Statutory Instrument 77 of 2024. The applied rate is twice the current month’s average less the prior month’s, a momentum-doubling formula. The regulation names no series; the reconstruction identified the Bank’s retail average selling rate, which reproduces every applied rate to within four ngwee. “Behind the Petrol and Diesel Pump Price” (July 2026) set out the mechanism. Its October build was 20.21 on the complete September dailies; the ERB applied 20.20. The Board’s October statement separately cites the kwacha slipping from 19.28 to 19.76 over the review period: the market’s move in the window, not the slate it applied.

What the suspension cost, and what the 30 September decision did with it, is a first-order fiscal quantity (Table 11).

The reconstruction in Table 11 puts the cost at ZMW7.47 to 8.42bn, about USD400m at the year’s rates. The Economics Association of Zambia carries the same figure. At the pre-restoration run rate the suspension forgave roughly 7 per cent of the year’s domestic revenue target, annualised. The IMF counts it among the drivers that cut the projected primary surplus from 3.8 to 1.1 per cent of GDP. The second layer changed sides across the year. The line charged in February and March, ZMW3,330 to 3,640 per cubic metre at February’s peak, ZMW3.33 to 3.64 a litre, and again in June and July. It paid out in April and May, and again from August. Diesel litres funded 86 per cent of the scheme’s standing balance at its August peak. By end September its reconstructed room was below one month of a full pin.

The 30 September review resolved all three layers together, and Table 12 grades the detail. Excise came back, with VAT to follow on 1 December. The stabilisation line printed zero on every product for the first month since it began. The pin broke, and pumps reset a quarter higher at midnight. Table 11A sets out what the staging recovered and what it left.

Two consequences carry forward. The headline says taxes are back, but excise is 16 to 18 per cent of the monthly wedge, and the VAT remainder in Table 11A stays forgone through October and November. Its return on 1 December is a pre-announced pump rise of roughly 16 per cent, about ZMW5 a litre, unless something absorbs it; the conserved stabilisation line could, and Table 11A’s note prices the room. And the November Committee meets on 25 and 26 November, before that reinstatement, so the next rate decision will also be taken on a partially suspended index.

A third consequence sits in the fund. It did not spend in October. If it spends in December instead, pump prices need not rise as much when the VAT returns, or at all, depending on where crude and the crack spreads have moved by then. The November build-up and the 1 December prices will say.

Two things stay true either way. A cushioned rise is postponed, not cancelled, and the money spent cushioning is the money a Gulf or El Niño shock would need. The tax cost does not disappear either: Table 11’s ZMW7.47 to 8.42bn already sunk through September, plus October and November’s VAT line, stands whatever December does. A December draw moves the cost from forgone revenue to the fund’s ledger, financed by the windfall-month payments motorists already made.

It would be the restore-and-pin of “Behind the Petrol and Diesel Pump Price”, staged. Today the instrument is a single line in the wholesale build-up, as that essay showed, not a standing fund. Built properly it needs transparent rules, quarterly holding, and pre-funding at a formal launch, because a December draw spends the conserved balance and the fund would start empty. The case is monetary as much as fiscal: a rule-based fund smooths the pump path the CPI reads, instead of handing the index a quarter-sized step. The precedent is regional and live: South Africa runs the formal version, formula errors amortised through a self-adjusting levy under a published rule, the balance printed monthly.

If that is the plan, section 5’s reading follows: the cut was priced off a December the authorities intend to engineer, coordination in fact and balance-sheet management in mandate. The upside risks stand; the recommendation is unchanged.

The staging has an index arithmetic, and it defines this note’s full-tax counterfactual: the inflation rate as it would print with the fuel taxes restored. It is the index section 3 stress-tests the 20-year against. The chain runs forward from published numbers. On the ERB build-ups the suspended taxes are about a fifth of the full-tax pump price, so restoring them raises the paid price by about a quarter. The published Transport weight is 5.8 per cent; fuel at an effective 5 inside it turns the one-quarter price gap into a 1.3 per cent index gap, 1.075 against 1.061. On September’s printed 6.1 that is a counterfactual near 7.5. The wedge converges: about 1.4 points in September, 0.8 from October with only the VAT missing, none from 1 December, when print and counterfactual merge. On that weight, held fixed, the October reset adds roughly 1.2 percentage points in the month it enters and the December reinstatement roughly 0.8 more, before second-round effects. Both figures are graded when the October 2026 and January 2027 bulletins print.

Base effects run the other side of the ledger. October 2025’s 0.4 drops out of the October comparison, and December 2025’s 1.5, the largest base month left in the twelve-month rate, drops out exactly when the VAT step lands. The base is not all cover: Decembers run seasonally warm, 0.8, 1.0 and 1.2 in the three years before 2025, so December 2026 brings its own print. The trend shrinks it: 2026 prints average under half their 2025 counterparts, so the step can enter the price level and leave the year-on-year rate near the band’s floor.

That arithmetic is what the Bank’s 6.7 average for 2026 and 6.0 for 2027 lean on, and the annual averages still need the rest of the index to behave; the price level rises either way. The staging matches the calendar: the smaller step, the excise, lands first; the larger, the VAT, lands in the month with the most base cover. As sequencing, the December choice is well judged.

The cover then expires. Both steps stay in the twelve-month rate deep into 2027, the 2025 bases are gone by January, and the 6.0 for 2027 must carry them. The first half of 2026 ran on the opposite effect: the drought-era prints of early 2025, 2.1 in January and 2.4 in February, fell out of the rate. The record harvest then pulled food inflation from near thirteen to six. From January 2027 the bases flip. The 2026 prints hand over 0.5 to 0.7, the planting months are seasonally the year’s warmest, and the VAT step, El Niño and Gulf crude all land inside them. December’s cover is the calendar’s last; the months after it carry none.

The import bill is already moving on Middle East crude, per the ERB’s own October citation. The outlook’s other pillar is maize: the Committee’s statement rests its favourable path on stable grain prices, and section 2.6’s four million tonnes is that pillar’s balance sheet, the grain twin of the fuel fund.

Bank of Zambia Working Paper WP/2026/2, the source of this note’s epigraph, finds expectations broadly unanchored and adaptive, with the exchange rate and the policy rate their key drivers. An adaptive public does not read the slate; it reads the pump prices the slate sets, and the slate moved: September’s average was 19.67 and the October setting applies 20.20.

There is a fiscal ledger on the exchange rate, and it cuts both ways. The budget carries roughly ZMW21.7bn of external interest and principal, ZMW11.58bn of it remaining after July; every 1 per cent the kwacha loses adds about ZMW116m to that remainder. Import VAT at ZMW42.5bn is charged on the kwacha value of imports, and royalty at ZMW18.2bn and mining company tax at ZMW13.2bn convert dollar receipts. From the build’s 20.21 to 22 at constant imports, the import-VAT line alone gains roughly ZMW1.5bn over the five remaining months. That is a gross revenue sensitivity, not the net fiscal effect, and the cost lands in the price level the Bank targets.

The house corridor stands: 18 to 20 suited stabilisation through the vote; 20 to 22 is the current drift; 22 to 24 is the medium term. Two rates move inside it, and they are different objects. The retail series the slate runs on averaged 19.67 in September; the slate, doubling momentum by construction, already applies a 20 handle for October. Any further market weakness reaches the pumps amplified, and the working paper names the exchange rate among the two key drivers of expectations. That is the bind the fuel decision tightened. With the taxes back and adding their 1.2 points, the kwacha is the lever that still defends the index. It defends it from the strongest real effective level since 2003 on the Bank’s readings, a third below the long-run average. Holding the level defends the CPI and forgoes the revenue; drifting collects the revenue and hands the index the slate’s doubled version of the move.

5. Wednesday, Priced in Advance

The matrix was fixed before the 30 September announcements and the outturn column completed after them; nothing in the expectation column has been revised.

The case for the cut and the critique of it share the same three facts. At 10.75 the real rate is positive on either index: 4.65 on the printed 6.1, near 3 on the 7.5 counterfactual. Kwacha private credit, at 7.0 per cent of GDP, gives the cut little credit inflation to generate; its most direct effect is the sovereign’s rollover cost into December’s wall. And on the Bank’s own working paper, the exchange rate stands beside the policy rate as a key driver of expectations. Read one way, that is room to ease. Read the other way, a cut whose liquidity reaches government paper before private credit is fiscal accommodation by function, whatever its label. It arrived at five to ten times the matrix’s forecast, in the quarter the financing provision runs out.

The paper’s coefficients price the week: the 250 points add roughly half a point to the gap between expectations and target; the month’s exchange rate move, about 0.03. The working is the paper’s own gloss, 0.01 on the deviation per 1 per cent exchange rate move and 0.190 per policy point, fit on quarterly professional forecasts through 2024. These are model sensitivities, not measured announcement effects, and they are the second-round channel section 4’s 1.2 and 0.8 exclude. Even a full move to the drift corridor’s 22, about 13 per cent from the end-September interbank, prices near 0.13 on the same coefficient: a quarter of the cut’s half point. On the paper’s own arithmetic, the week’s larger de-anchoring impulse was the cut, not the kwacha.

The Committee cut “taking into account the lower inflation outcome, the projection that inflation will remain within the target band over the forecast horizon, and the identified upside risks to the inflation outlook”. The same document calls the decision “supportive of the national growth agenda”, under the Act’s price stability mandate. On this note’s reading, those identified risks are underpriced. The counterfactual already runs near 7.5, section 4’s calendar hands both tax steps to 2027 with no cover left, and the cut landed ahead of the inflation risk, not behind it. Twelve hours later, the fiscal authorities reset the index the Committee had eased against. No surprise reached the easer: the expiry and the review date sat on the calendar it met against, and October’s 1.2 points were computable at the table. A committee that cuts the night before a reset it can compute has priced the reset. Its projections say the same: a 6.7 average of the monthly rates needs the fourth quarter at 5.9, since January to September’s prints sum to 62.7. The October reset alone puts that quarter near 7 on this note’s weight. Either the August round, reaffirmed the evening the fuel decision landed, was stale at the cut, or the December pin is what the Bank’s own average needs: section 4’s coordination made quantitative. The October bulletin grades the branch. The decision reads better as balance-sheet management than as inflation targeting, and the October print, the November meeting and the December reinstatement will say which reading carries more of the truth.

The forecast missed and the matrix prints it; the recommendation was another thing. A hold, with the reserve relief re-issued in its published conditional form, earned against private lending. The auctions had already taken the curve down without the Committee’s help; a hold cost little and kept credibility for a November meeting that will carry October’s reset in its data. The structural objection fits in one line: credit access does not widen while freed liquidity channels to government paper.

One development could re-anchor all of it. The government is negotiating a successor arrangement with the IMF: the request Situmbeko Musokotwane, reappointed Finance Minister on 14 September, put on the record in July. A new round of talks was announced on 30 September, as this note closed. The sequence matters. The original 2.1 per cent fiscal deficit, the borrowing ceilings and the primary-surplus path all date from the ECF, and the supplementary that rewrote them arrived after that programme ended. Most of what this note documents is a fiscal year with the anchor removed. A concluded programme restores the anchor as a performance criterion rather than a repealable number, restores the data discipline on exactly the publications this note found defective, and reopens the external financing door. Nothing guarantees the negotiation concludes; the fourth-quarter notes track it alongside the arithmetic.

In five weeks of midwinter, the Republic paid USD1.12bn abroad at 82 to 84 cents on the dollar to insure against its own recovery. At home, over the same months, it rationed pensions and school feeding, spent its deposit buffer in the month before the vote, and ran its revised fiscal deficit ceiling to three quarters gone. The external act was financed, publicised and graded a success. The domestic ledger was published with its signs dropped, scored against a superseded budget, and its reserves cost surfaced 66 days later in a statistical annex. The same state sets the auction cut-offs, the publication calendar and, this week, the policy rate, against an index whose floor its own Treasury had set. The slower verifiers in Table 12, the August MEI, the August reserves and the October credit data, are graded in the October note, exactly as Table 2 grades June’s.

Sources

Bank of Uganda, Treasury bond auction results (Kampala, 9 September 2026) Bank of Zambia, Circular No. 10 of 2026 and Gazette Notice No. 1243 (Lusaka, 31 July 2026) Bank of Zambia, Commercial Banks’ Liquidity fortnightly (Lusaka, August 2026) Bank of Zambia, Daily Exchange Rates to 30 September 2026 (Lusaka, 2026) Bank of Zambia, Fortnightly Statistics (Lusaka, 4 September 2026) Bank of Zambia, Fortnightly Time Series ending 18 September 2026 (Lusaka, 2026) Bank of Zambia, Governor’s Media Presentation, Q3 2026 (Lusaka, 30 September 2026) Bank of Zambia, Monetary Policy Committee Statement for Q3 2026 (Lusaka, 30 September 2026) Bank of Zambia, Public Notice: Issuance of Quarter Four Calendar and the Introduction of the 20-Year Government Bond (Lusaka, 2026) Bank of Zambia, Tender invitations and result sheets, Treasury bills 01/2026 to 19/2026 and bonds 01/2026/BA to 07/2026/BA (Lusaka, 2026) Bank of Zambia, Working Paper WP/2026/2, Are Inflation Expectations in Zambia Anchored? What are the Drivers? (Lusaka, June 2026) Bloomberg composite pricing (CBBT), the Republic of Zambia’s 2033 restructuring bond (1 October 2026) Canary Compass, After the Count (Lusaka, August 2026) Canary Compass, Behind the Petrol and Diesel Pump Price (Lusaka, July 2026) Canary Compass, Copper Output and the 2026 Royalty Arithmetic (Lusaka, January 2026) Canary Compass, Growth Without Diffusion (Lusaka, March 2026) Canary Compass, Seven Stars That Refuse to Align (Lusaka, May 2026) Canary Compass, The Most Expensive Recovery (Lusaka, June 2026) Canary Compass, The Year as a Single Equation, private note (Lusaka, July 2026) Canary Compass, Zambia 2026 Domestic Market Absorption: The Arithmetic Through End-January (Lusaka, 2026) Canary Compass, Zambia Monetary Policy, Part 3: Foundations for Transmission (Lusaka, December 2025) Central Bank of Kenya, Treasury bond auction results (Nairobi, 21 September 2026) Central Bank of Nigeria, open market operations auction results (Abuja, 29 September 2026) CSPR, Press Statement on Mid-Year Budget Performance and Summary Analysis of the 2026 Supplementary Budget (Lusaka, 2026) Debt Management Office Nigeria, FGN bond auction results (Abuja, September 2026) Ecofin Agency, coverage of the Food Reserve Agency board statement (September 2026) Economics Association of Zambia, Policy Reaction Statement on the Monetary Policy Committee Decision (Lusaka, 30 September 2026) Energy Regulation Board, Monthly Wholesale and Pump Price Build-Ups, December 2025 to October 2026, and 2025 Statistical Bulletin (Lusaka, 2026) Energy Regulation Board, Press Release: Revision of Petroleum Pump Prices for October 2026 (Lusaka, 30 September 2026) Food Reserve Agency, 2025 crop marketing season final purchases, per the parliamentary record and official reporting (Lusaka, 2026) Food Reserve Agency, Board Statement on the 2026 Crop Marketing Exercise (Lusaka, 29 September 2026) Government of the Republic of Zambia, Statutory Instrument No. 77 of 2024, The Energy Regulation (Petroleum Products Price Setting) Regulations (Lusaka, December 2024) IMF, Staff Statement on Zambia (Washington DC, May 2026) Kenya National Bureau of Statistics, Consumer Price Index, September 2026 (Nairobi, 2026) Ministry of Finance and National Planning, 2026 Annual Borrowing Plan (Lusaka, 2025) Ministry of Finance and National Planning, 2026 Budget Address (Lusaka, 26 September 2025) Ministry of Finance and National Planning, Debt Statistical Bulletin Q4 2025, Table 27 (Lusaka, 2026) Ministry of Finance and National Planning, Monthly Economic Indicators, January to July 2026 (Lusaka, 2026) Ministry of Finance and National Planning, Supplementary Estimates of Expenditure No. 1 of 2026, as transcribed in ZIPAR and UNICEF (Lusaka, 2026) National Assembly of Zambia, The Supplementary Appropriation (2026) Act, No. 78 of 2026 (Lusaka, assented 4 June 2026) NOAA Climate Prediction Center, ENSO outlook (Washington DC, August 2026) Press reporting of the 14 September 2026 cabinet appointments (Lusaka, September 2026) Press reporting of the Agriculture Minister’s January 2026 statement of the 2.5m-tonne strategic reserve target (Lusaka, 2026) Press reporting of the Agriculture Minister’s May 2026 statement to Parliament on national maize storage capacity (Lusaka, 2026) Press reporting of the Food Reserve Agency’s 540,000-tonne maize export agreement with Kenya’s Baita Trading, signed 9 September 2026 (Lusaka, September 2026) Press reporting of the Bank of Zambia Governor’s 7 July 2026 remarks on reserves (Lusaka, July 2026) Press reporting of the FRA executive director’s 18 September 2026 statement on maize holdings and storage (Lusaka, September 2026) Press reporting of the government’s 30 September 2026 announcement of successor-programme discussions with an IMF mission (Lusaka, October 2026) Press reporting of the kwacha’s inclusion in JP Morgan’s new frontier local-currency bond index (September 2026) Press reporting of the Minister of Finance’s supplementary budget presentation to the National Assembly, including its financing sources (Lusaka, May 2026) rbond.co.za, South African government bond mark-to-market yields (30 September 2026) Republic of Zambia, Tender Offer announcements for the 2053 notes: Amendment and Extension, Early Results and Final Results, RNS (London, 4, 10 and 12 June 2026) Reuters, record of Finance Minister Situmbeko Musokotwane’s programme request (July 2026) Secretary to the Treasury, April 2026 presentation on agricultural arrears, as reported (Lusaka, 2026) Statistics South Africa, Consumer Price Index, July 2026 (Pretoria, 2026) Times of Zambia, “FRA maize purchases hit record K20bn” (Lusaka, 30 September 2026) Uganda Bureau of Statistics, Consumer Price Index, September 2026 (Kampala, 2026) ZAMACE (Zambian Commodity Exchange), market reports on maize offer and regional prices (Lusaka, September 2026) ZamStats, The Monthly, Volume 249 (Lusaka, December 2023), and Monthly Inflation, December 2025 release (Lusaka, December 2025) ZamStats, The Monthly, Volume 276, preliminary 2025 GDP estimates (Lusaka, March 2026) ZamStats, The Monthly, Volume 282 (Lusaka, 25 September 2026) ZIPAR and UNICEF, Mid-Year Budget Performance Analysis Zambia 2026 (Lusaka, August 2026)

All calculations are available for inspection with full workings.

Canary Compass. Structure Before Sentiment.

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: The difference between a mark and a model is order and witness.

In 2008 I took the graduate seat on a trading desk, and the first responsibility they handed me was funding. The desk ran positions across African markets; funding ran both ways. Where we held local currency, my job was to lend it well. Where a position needed local currency we lacked, my job was to borrow it cheaply. Lend too cheap and the money earns less than it could. Borrow too dear and the trade gives back more profit than it should. The profit that counted was the net, the position and its funding taken together. Funding was the desk’s first schoolroom. Once funding made sense, the rest came down to one question: how long to borrow or lend. The answer lived in the position and the price of each length of money.

I arrived knowing none of this. Early on, a senior trader told me to call one of the banks and ask for Tom Next. I raised the bank on the Reuters dealing machine and typed, politely, asking to speak to Tom Next. Tom next is not a person. It is one of the shortest borrowings there is: money taken tomorrow, returned the day after. My request sat there in writing, and the floor enjoyed it.

The joke cost me nothing but burning ears. The mistake that followed cost more. The desk was long dollars, a trade that earns when the local currency weakens, funded by borrowing the local currency, and the borrowing was mine. The need was daily, at the short end of the curve where money is cheap, and African curves tend to be steep: the longer the borrowing, the dearer the money. Trying to be efficient, and a little clever, I borrowed for a whole year, without asking how long the position needed the money. No loss appeared on the day. The cost arrived as a bleed: every morning the book paid the year’s rate for money it needed a day at a time, eating the profit the falling currency was making. The way out was to lend the year back out, in a market whose spread charges heavily for changing your mind, and the loss was taken at once. I remember the stern looks. In eighteen years of trading I have never been as scared, before or since, as I was then, a graduate who had tied the book to a year of expensive money on his own initiative.

I never made that mistake again. The cost sat in the book in plain view, with my name on it and a date. That is what a desk is, a place where being wrong costs something, quickly and in your own account. The desk charged positions, never opinions.

An argument is a position nobody marks.

Conversations do get marked; friends will say so, and they are half right. Mockery, or the group that quietly stops inviting you: each is a real price, and it prices your distance from the room. Your distance from the truth goes unbilled. A man wrong alongside the whole table is paid in belonging, and a man right against the table pays. Enron marked its contracts to a model of its own, booking projected profits as real. A room does the same with its opinions, and the model it marks to is the one that keeps the room intact. Every opinion inside a price has money behind it, so the market’s mistakes settle, whether or not the room comes along.

Hold a wrong view for twenty years and the truth charges you nothing. Nobody walks over with the stern look. I have watched economists stay certain about an economy for a decade, wrong at every point where a checkpoint should have stood, at no cost, because none was ever agreed. I have watched public forecasts miss and the forecaster grow more confident, refinanced by hindsight, since yesterday’s miss never rolls into today’s cost. Our longest debates run long for exactly this reason.

Marking to market is mercy dressed as discipline. It shows the loss while it is young and keeps open the chance to trade out of it and remain alive. The mercy is not gentleness. Months after my lesson, 2008 turned savage and marked every book in the world, some so violently that the book did not survive. An unmarked view keeps no calendar. It can be held forever, or it settles once, when life calls it in, the loss arriving whole, no time left to adjust.

Mark your own book. Before an argument, write down the result or the date that would close your position, and ask the person across the table for theirs. Argue values as long as you like, but a claim about the world owes a close. If neither of you can name one, you are rehearsing, not arguing. The difference between a mark and a model is order and witness. The condition is set before the position, and a copy sits in someone else’s hands, where it cannot be narrated away. That is all a wager has ever been.

Last week’s essay came out of a debate I have been in with friends for years. My own side of it has long been written and dated. The exchange is what has never happened. No close of mine has ever crossed that table, and I never asked for theirs. We have argued in a market with no prices, each of us marking to his own model and calling it honest.

The graduate bought eighteen years of discipline with one week of fear. Cheap tuition. No bill has ever reached the views I take to that table, and that is the warning. The admission now sits in other hands, including the hands I argue with. The next argument starts from it. That desk is gone. The book is still mine to keep.

AI-illustration: There is no one to fire.

For years now the same argument has kept finding my table. It arrives with friends, over food or over social media, and it always starts as a question: which system actually works, communism, socialism, libertarianism, capitalism? Nobody asks it idly. People feel locked out of the system they were told to believe in. Populism is rising on both flanks of it, and young people in the richest country on earth are embracing socialism in numbers their parents would not recognise. Sooner or later someone says it: capitalism has failed. My answer has been consistent for years, and I should have written this reflection long ago. Capitalism has not failed. What has been running in its name is crony capitalism.

I should declare where I stand. I call myself a disciplined libertarian communitarian. I believe in the individual, in enterprise, and in market prices as the most effective decentralised allocation mechanism humanity has found. I hold markets sacred but not sovereign, built to serve people rather than master them. I believe people in communities locked out for generations deserve honest, deliberate pathways to skills, capital and opportunity, not outcomes awarded by identity but access through which effort and merit can operate. Those pathways must be time-bound and audited, so that repair never calcifies into dependency. And I believe the state belongs in markets chiefly as their engineer and their referee, not as their permanent allocator. It lays the rails and keeps the registers; it polices the evidence and enforces the contracts. Where sovereignty demands that it do more than lay the rails, its intervention must be temporary, measurable and accountable, and then it must withdraw, because a state that lingers as a competitor becomes a parasite.

Crony capitalism is usually described from the top: the tender that finds the president’s friend, the bailout that finds the connected bank. All real. But that description misses where the mechanism lives for most people, which is not in State House. It lives in the plumbing, in the ordinary prudential machinery of finance, and it lives there without anyone intending it. The forms differ by country; the rule inside them does not, and it reads the same at a counter in Nairobi as it does in New York. Before any institution lends to you, or even banks you, it must answer two questions: who are you, and can you be trusted with money. The machinery built to answer them is KYC and risk assessment. Both exist for unimpeachable reasons. Laundering and terrorist financing are real, and so are credit losses. And both allocate access by network.

Take account opening first, because it is the door to everything else. Within living memory, and in some places still, opening an account required references: existing customers of the bank writing to vouch for you. Some countries have retired the rule. The rule never asked who you are; it asked who will stand for you. Those are different questions with different losers. Identity can be established by anyone with a document and a face. Sponsorship can only be produced by the already banked. And the rule’s quieter descendants are still on the forms. Proof of address wants a utility bill, a lease or a bank statement, and the circle closes: a bank statement proves an address for someone who already banks. The bill assumes a meter in your own name, which the settlement and the homestead never had. A wife whose lights run in her husband’s name can bridge the gap with a marriage certificate, one more paper that formal membership prints. The woman in a come-we-stay union has no such paper; the wedding that would have papered her was never held, often because it could not be afforded. Her home is real in every way except the one the form can read.

Source of funds asks a fair question, where does your money come from, then accepts only the answers formal income can print: payslips, contracts, sale agreements. The trader who earns honestly in cash every day is not unbanked; his phone holds an account, and the wall stands where the amounts grow. Bank surveys keep recording his rejection as incomplete financial records, which is exact in the cruellest way: the records are complete for his life and incomplete for the form. Some banks bridge it honestly, sending a loan officer to the stall to count stock and watch the till, the institution paying the verification cost itself instead of billing it to his paperwork. The visit is expensive per customer, so it stays the exception; the paper list stays the rule. None of these requirements is a price. There is no amount the unsponsored, the unaddressed or the undocumented can hand across the counter to satisfy them. A price rations by willingness to pay. These rules ration by prior membership. Whatever we call that, we should stop calling it a market mechanism. Screening itself is a market’s honest answer to being lied to; the indictment is narrower. When every admissible proof is manufactured by prior membership, screening stops sorting risk and starts sorting insiders.

The credit side repeats the shape at every gate. A loan wants a guarantor, and the guarantor must himself be creditworthy. Collateral must carry title, wealth you must already hold before capital will reach you. A credit history can only be built with credit, the one thing the applicant came to get. Payslips and statements assume formal employment and the account whose door we just examined. And before any of it, a risk rating has often scored the applicant’s occupation, postcode and cash habits as reasons for caution. None of these rules was written to exclude. I know this from the drafting side of the table, where every one of them is written, sincerely, as prudence. I also know that my own accounts open on a phone call or an email, because the network already knows me, and that the corridor has never once slowed me down. Walked end to end, the rules compound into something no one designed: a corridor where capital flows cheapest to those nearest it, and money grows dearer with every step of distance from the network. The same rule runs upward: the network that prices the distant out lets the connected default and refinance, cronyism at the top and exclusion at the bottom being one rule applied to different people. Underneath all of it runs one law. The questions are fair. Who are you, where does your money come from, can you repay: every institution is entitled to ask them. The gate is never the question. The gate is the list of admissible evidence, because every list is written so that only the already-inside can produce what is on it.

Communism won. I mean something narrow and checkable by won. The ideology lost, and the planned economy lost with it; what survived is the technology communism ran openly, sponsorship. The method is older than both ideologies: the letter of introduction was opening bank doors long before the Soviet Union existed. Communist systems made it an organising practice of everyday access, the party member who vouched for you, the character letter from a cadre in good standing. And the system that defeated communism kept the method running quietly, wearing prudence, while claiming price had won. The front of the house takes money; the door is opened by names. Crony capitalism sits one mirror away from the capitalism we keep defending in those debates, and the glass hangs in the back office, not in State House. This is cronyism without a crony, and that is the harder kind, because there is no one to fire.

Even the writers of the global rulebook now concede the effect. The Financial Action Task Force, the global standard-setter for bank compliance, has acknowledged since 2011 that overcautious application of its own safeguards shuts legitimate customers out of the formal system. It has spent this decade rewriting its guidance to say so. The scale is not small: the World Bank’s newest count still finds 1.3 billion adults holding no account anywhere, the mobile wallet counted, and missing documents remain among the barriers they name. The serious defender of the current rules deserves an answer on their strongest ground. Cash-intensive, undocumented customers are scored as higher laundering risk as a class, and the system’s concession is the simplified tier, lawful because it is capped. But a class score is the cheap proxy this machinery reaches for wherever individual verification is dear, and the caps hold access at token amounts, so the corridor reappears at every tier boundary. So the honest question is how to verify without handing the door to the network, because the losses and the laundering are real. And the world has already run the experiments.

The oldest is the exam number. Universities learned hard lessons about examiners and surnames, and answered by making the candidate a number, so that the script is judged and the person is invisible. A script at least contains everything the grade requires. Even there the person leaked: an examiner could still hear, in the fluency of the prose, which schooling had trained the hand. The number hides the name; the writing keeps speaking. Then the harder lesson. In 1997 a computer scientist named Latanya Sweeney took a supposedly anonymised release of medical records and re-identified the sitting governor of Massachusetts from three innocent facts, postal code, date of birth and sex. She went on to estimate that up to 87 per cent of the population of the United States is unique in those three facts. Strip the name and the person remains legible. An index has to be far cleverer than a deleted field, and any counter that promises to judge the record while blind to the person inherits that burden. Then the hardest lesson. France took the experiment national: its employment service anonymised CVs to strip origin and neighbourhood from hiring. Among the firms that took part, minority candidates did worse under anonymity, interviewed less and hired less, because blindness had also deleted the context that let a fair reader weigh a thin CV honestly. The state abandoned the policy. Hiding identity is not enough, and done naively it runs backwards.

The design built from all three lessons is disclosure of claims without disclosure of identity. Prove the repayment record, the income, the standing; keep the name, the tribe and the address out of the underwriting room. The design escapes France’s mistake by substitution: France subtracted information and added none, while this design trades a thinner identity for a thicker record, so the fair reader is given more to weigh, not less. That design confines identity rather than abolishing it: the compliance desk verifies the person, the underwriting desk receives the verified claims, and the pedigree crosses into neither decision. The split answers for the person and the history; it does not claim to trace every shilling’s pedigree. And the design leaves a dependency standing: someone must write the first record for the person the system has never met.

Proof that such evidence can be manufactured outside the formal system exists at scale. Grameen lent to Bangladesh’s poorest women on joint liability, the group standing for each member, local knowledge doing the underwriting, and repayment held. In 2002 the bank dropped the group guarantee, though the groups kept meeting. My reading: years of repayment had given each member a record of her own, and the record had become evidence enough. On that reading, the group was never only a loan scheme; it was a factory for admissible evidence, and its members graduated out of needing it. Once the record can travel, identity can stay home. The village knows everything; the formal counter needs only what the record proves. The SACCO, the chama and the village bank do the same work here. The community stakes its own money on its member, which is more than any reference letter ever risked, and every cycle writes another line of record.

Banks have noticed the chamas, and they have incorporated exactly the wrong half. They court the group’s deposits and lend against its pooled cash or a multiple of it, money anchored in money. The member’s ledger is the half our evidence lists do not admit. Five faithful years in a banked chama count for nothing when she stands at the counter alone. The group’s money crossed; her record did not.

The proposal is the record’s own crossing, into her file, travelling with her. It has two doors, sight and word. Sight is the door India began opening nationwide, and it formalised nothing to do it. Group lending there had begun to sour, and the repair India chose was the record itself. Since 2016 its banks must file the borrowers of the larger group loans with the bureaus member by member, each woman’s name and her share of the debt written into a file of her own. The groups stayed informal by design; the lender’s own eyes make that record. And the same circular drew the boundary the chama already showed us: lending among members from their own savings was left out expressly, and each member’s own repayments were deferred to a later phase. The door is real, and it is half open. India’s microloans have carried bureau records since a crisis first forced them, and crisis has still returned where lenders raced past what the file showed. A record disciplines only the lender bound to read it.

Kenya’s mobile rail is the same door, the observer reporting what it sees. Sight is even being carried into the group itself, and Zambia shows it plainly. Its banks court village banking by name, and group platforms trade openly there, member statements, automated share-outs, the whole cycle running where the host can see each member. It is still sight: the record lives on the observer’s own rails. The host already files any loan it makes to her, because lenders must report what they lend. But the cycle itself, the saving and internal repaying the platform merely watches, is not the host’s loan, so no mandate files it. That record stays inside the host’s walls until a rule makes the observer file what it sees, the rule India wrote for the loan and deferred for the cycle. It reaches only the groups that move their books in; the notebooks that stay home need the second door.

Word is the second door, for the ledger only the group sees, and a word must be answerable. The formal worker has always had this door; the employer’s letter is testimony the counter accepts, a letter that stakes nothing, admitted on formal membership alone. Even Kenya’s licensed SACCOs, when they report what they lend, are sight, the lender filing its own loans. The group’s word, its testimony to the saving and standing only it sees, has no such door: no due-diligence or bureau rule I can find treats the unlent cycle as the employer’s letter is treated. Some markets already file what she pays the power company and the landlord; what she pays her circle is filed by no one.

The proposal writes the collective the letterhead the employer has always held. The group that testifies keeps books and accountable trustees on a registry, and it stakes what it signs: answerable for the truth of the record, never for a member’s future debts. The registry reads the collective’s books, capital and stake; it does not march its trustees back through the old paperwork, because who they are was always the easy half of the question. Formalisation is asked only at this door, priced to the attesting function: a toll, not the point, because the point is admissibility. And at the word door the toll pays the group back: once its word is admissible, registering is what turns the five faithful years only it could see into years that count. Warehouse-receipt statutes, established across this continent, turn a farmer’s stored maize into bankable collateral instead of invisible wealth. The same principle holds at every scale: do not lower the standard, widen what counts. The two doors do not reach everyone; they reach the millions the current gate pretends do not exist.

The admissible record has a second buyer waiting behind the first. A loan book with a filed loss history can be rated and sold: Bangladesh securitised micro-loans two decades ago, and India’s joint-liability paper now trades at scale. This continent can already securitise a recorded book; it cannot securitise an unrecorded circle, because capital markets cannot price what no record shows. The debts themselves are the asset; the record is what lets a buyer price them. Open the record, and the same five faithful years underwrite twice, evidence at the counter now and, in time, the loss history beneath paper an investor can hold. From admissible evidence to investable asset, the record is the precondition, and the rest is machinery markets know how to build.

A design that replaces a gate must be tested harder than the gate it replaces. It can fail five ways, each visible from here, and each needs engineering against from the first day. The first failure is that the cooperative is also a network. The same intimacy that makes its information good can blackball the newcomer, the wrong clan or the unmarried woman. So the group’s word enters as one admissible route among several, never the only door. The second is fraud. The moment standing counts, standing will be manufactured: repayment rings farmed for records, pyramids in cooperative clothing. Kenya has met both. Against them stands the design’s one law: testimony counts only where the testifier stakes money or licence on its truth. The referee’s letter cost the referee nothing, which is how we got here. The guarantor stakes money but is network-gated. The regulated cooperative stakes both and is reachable.

The third failure is surveillance. A lender that may read everything is a new gate wearing a friendlier face. So the record travels by consent, and no wider than the decision needs. The fourth is France’s lesson: removing information can move bias instead of removing it. So the design is judged on results named in advance, who came in and what the losses did, never on its intentions. The fifth failure is the scaffold that never lets go, and here my own politics turns on my own prescription. Pathways built for repair must be built to expire, but the clock is the record, not the calendar. A pathway ends for each person on the day her own file can answer the question it answered for her, the way the group guarantee ended in the Grameen reading above. The pathway itself stays open, because new people arrive at the counter every day. The evidence rules face a different discipline: no expiry, periodic review, the regulator that wrote them returning to ask who they actually let in and what it cost.

Nothing above waits on an invention. Credentials that prove a claim while withholding the rest run today. The mobile driving licence in more than twenty American states can show a bartender that its holder is of age without showing the address on the card. National registers are learning to answer who you are in seconds. Kenya’s rail, met above as sight, turned millions of daily transactions into scorable evidence and wrote repayment records at national scale, no referee and no collateral anywhere. Kenya also ran the failures first, blacklist-only reporting that punished without accrediting, and apps that harvested contacts and messages instead of indexing, and the corrections to both were admissibility rules. The cooperatives are manufacturing the records, and registries for movable collateral are statute across the continent. What is missing is admissibility: rule changes of exactly the kind parliaments and regulators pass every year, a line amended in a prudential guideline, an evidence list widened.

The evidence list has survived for a reason worth naming: the officer who admits one mule is punished, and the officer who refuses a hundred traders is not. So every desk rationally over-verifies, and the caution hardens into the forms. It stays hardened until the scoring changes. The evaluation that punishes the missed mule must also score the refused trader. Then the gate opens the way Kenya’s payments gate opened. One regulated entrant discovers the excluded are a market. Once its book is proven, the incumbents follow it in. But the entrant must sit inside the same penalty regime. Otherwise the incumbents will applaud the pilot and keep the old list. That is what makes this practical rather than utopian. It is also what makes the failure to do it a choice.

Financial exclusion has been filed for decades under charity, a welfare problem, someone else’s department. It belongs under the corruption of capitalism, a competition problem, owed a fix by everyone who claims to defend markets. A system that allocates capital by network while calling itself a market is not free; it is captured, and it is corrupting itself. And the first amendments are known. One is the due-diligence paragraph listing acceptable proof of address and source, opened to the regulated cooperative’s attestation. The other is the filing rule, so that the cycle a platform watches or a group attests writes into her own file and travels into underwriting. That is financing architecture Africans can enter at the level of the person, and not only at the level of the sovereign. The rescue of capitalism from its crony twin begins at the counter, not in State House. It ends in those two paragraphs, on the day the forms ask who you are and what your record shows, instead of who you know.

AI-illustration: 15.04 per cent to 2040: the bond behind Kenya's hosting decade.

Sometime today a young man in Nairobi will deposit two hundred shillings into a betting account. Ten shillings of excise will arise on that deposit, under the 5 per cent deposit tax Parliament introduced in mid-2025. The excise joins the pooled collections of the Kenya Revenue Authority (KRA); appropriations flow from there to the Sports Fund, and an escrowed portion services a special purpose trust called Linzi FinCo 003. In January the trust pays its noteholders again, and the largest of them are two public retirement funds, PSSF and NSSF. A portion of the pooled receipts will have travelled from betting wallets to retirement accounts, by way of a construction site on Ngong Road. There a 60,000-seat stadium, Talanta Sports City, approved for renaming as the Raila Odinga International Stadium at its official opening, is rising to host AFCON from June 2027. Across town, the older Kasarani stadium has just been named host of the 2029 World Athletics Championships, awarded by Tuesday’s World Athletics Council vote, the first ever held in Africa.

Award week set off an argument about whether Kenya deserved the Worlds. The argument missed the machine paying for the decade: Kenya finances its hosting ambitions by selling the future, and the next sale, KES38.74bn against the same Fund, is already in preparation. Two outcomes are possible. Arbitrage is value that lasts only while the debt stays off the official count; architecture is value that survives being counted.

1. The Machine

The Talanta financing closed on 30 June 2025. Linzi FinCo 003 Trust, arranged by Liaison Financial Services with KCB Investment Bank and CPF Capital, raised 44.79 billion Kenyan shillings (KES44.79bn) against future Sports Fund receipts at a subscription of 100.19 per cent. The established terms: 15 years to July 2040, an IRR of 15.04 per cent, 30 semi-annual redemptions beginning January 2026, interest tax-exempt like government infrastructure bonds, a GCR rating of AA(KE)(IR), and a listing on the Nairobi Securities Exchange’s restricted segment in KES100,000 multiples, a sub-segment that carries no public-disclosure obligation, so the offering memorandum circulates among qualifying investors only. Protections: assigned receivables, an escrow account, a reported three-month debt service reserve, and a standby letter of credit from KCB against delays in Treasury disbursement, an instructive choice of insured peril. Reported offer terms also include government safeguards against Sports Fund shortfalls, which sit uneasily beside the absence of a formal guarantee.

Start from what the public actually has, set out in Table 1.

Ordinary bond quotation reads 15.04 as compounded semi-annually, NACS in market shorthand: 7.52 per cent per half-year, KES3.37bn of interest on the full principal, and no day-count convention closes the gap to the KES3.25bn observed. The payments instead sit on 7.26 per cent per half-year, a 14.51 per cent NACS rate, which is identically 15.04 compounded annually, NACA. So the observed cash reads the issuer’s 15.04 as the annually compounded quote: on it, the two payments match the interest to within rounding, and none of the principal has moved yet. In what follows, the ordinary reading takes 15.04 as NACS and the effective reading takes it as NACA. The distance from the KES3.25bn seen to any full instalment is the puzzle itself, and Table 2 tests every simple shape the bond could take.

The table is the argument, and the Fails-on column does the excluding; the bullet row’s KES97.5bn is, in passing, where the KES100bn-plus political claims come from. What the surviving pattern most resembles is a bond’s cash on a mortgage’s paperwork: money going out like fixed coupons on the annual quote of the rate, while the terms and every analyst model amortise on the other; no one has observed a KES3.80bn payment, only inferred it. And the Fund’s transfer need not equal the holders’ payout, because the reporting notes investor payouts include escrow investment income, so the gap between 3.25 and a full instalment may be bridged inside the trust, out of sight. Failing that, the triad closes. Lifetime interest of 57.6 requires principal repaid faster than level; transfers that match full interest, if they are the whole payment, mean principal repaid slower and more interest than level, not less. No single schedule squares the stated rate, the payments made and the reported total, so something else has to give: a stub period, fees inside the transfers, or the reporting itself. One document ends the argument, the repayment schedule; the issuer holds it and has not released it. Until then, every service number in this note is computed twice, on the two readings defined below. That a public bond held by public pension funds against a public levy must be modelled at all is the finding.

Bases and conventions, used throughout. The level basis is Table 2’s mortgage-style schedule on the ordinary reading: service of KES7.6bn a year, weighted life near ten years because equal payments repay principal late; equal chunks of principal instead would give 7.75 years. Its effective-reading twin, KES3.70bn per payment and KES7.4bn a year, sits between the two bases and is not carried separately. The reported basis doubles the KES3.25bn transfer: KES6.5bn a year. Which reading governs is, like the schedule itself, nowhere disclosed.

Two further facts from the memorandum, as reported, matter more than the arithmetic. First, the Fund’s statute runs out in 2028, 12 years before the notes mature; the memorandum as reported dates the sunset August, while the regulations’ 26 October 2018 commencement read with the Act’s ten-year rule points to October, and nothing public reconciles the two. If the Fund is not extended, that is an event of default. Commencing the Fund’s winding-up then triggers acceleration, all outstanding amounts immediately due and payable from a Fund whose reported assets stood at KES5.6bn in mid-2023, the notes holding a priority charge on them. That is thin against KES44.79bn, but more than ordinary public debt offers: unsecured sovereign paper meets acceleration with no ring-fenced assets at all, only the taxing power, the contrast Table 5’s security row draws. Second, the stadium itself is not collateral: the state holds it through Sports Kenya, and noteholders can reach only the receipts. If the Fund is ever wound up, the law says the Cabinet Secretary shall pay any deficit from the Exchequer, with the National Assembly’s approval: a statutory duty, behind a parliamentary gate. So the documents answer what the rhetoric leaves open: no guarantee in law, a residual in substance. The sunset, the default and acceleration provisions, the shortfall safeguards and the competing payment readings all come from the memorandum as reported, not documents in hand; each is flagged at first use, and later mentions inherit the flag.

The holders and the asset, from the reported record. Regulatory filings show PSSF at KES16.29bn and NSSF at KES7.9bn, 54 per cent of the issue between two public retirement schemes. The filings also name County Pension Fund at KES1.98bn, CPF Individual Pension Scheme at KES790.5m and the Local Authorities Pension Trust at KES197.7m, taking disclosed retirement-scheme holdings to about 61 per cent; every disclosed shilling is visible through the buyers’ regulator rather than the listing, and the residual is not publicly broken down. The works contract was signed at KES45.85bn, USD344.5m in the audit papers, up from an approved estimate near KES35bn; the government attributes the escalation to taxes, levies and import charges, and Parliament keeps questioning it with the Auditor-General’s findings on the table. The build stood at 68 per cent complete in February’s Budget Policy Statement and about 89 per cent by the May Senate hearing, with remaining proceeds directed to training facilities, pools and an indoor arena. The arithmetic does not close: a KES44.79bn raise, transaction costs, a KES45.85bn contract that alone exceeds it, and further scope besides. The offer material had projected the proceeds covering the stadium with room for the Ngong Road flyover and satellite pitches; the schedule bridging raise, costs, contract and scope is one more gap in the public record. A second facility of KES38.74bn, the sequel throughout this note, is being prepared against the same receipts for 33 county stadium projects, a transaction advisor under recruitment and a similar tenor expected.

None of this stands alone. The flagship of the family is the roads securitisation: KES7 of the KES25 fuel levy committed for ten years against a KES175bn target to clear contractor arrears, bridged by KES104bn from four banks, TDB, KCB, Absa and UBA, with UBA’s leg signed at USD150m. By December 2025 the Cabinet declared certified road arrears to end-2024 settled at KES123bn, with 875 contracts accelerated since April. The take-out bond remains unissued, its target cut to KES120bn by June 2026 with no disclosed timeline, while a further KES5 per litre has entered securitisation. The Railway Development Levy and part of the tourism levy are committed, a rail-levy tranche is planned for the standard-gauge railway extension toward Kisumu and Malaba, and press reporting puts the near-term pipeline toward a trillion shillings. National pending bills stood near KES470bn in early 2026, counties roughly KES160bn more per the Controller of Budget, so the practice has demand whatever the statisticians rule.

The fiscal frame, each figure dated. Kenya’s FY2025/26 plan projected a deficit of KES923.2bn, about 4.8 per cent of GDP, with net domestic borrowing of KES635.5bn; the enacted FY2026/27 budget carries a reported borrowing gap near KES1.15trn. Against that scale, the KES383bn the World Bank counts as raised from the three funds, these notes included, is real money: roughly three-fifths of one year’s domestic borrowing, accumulated across the deals. The Fund’s own money comes in three counts that do not describe the same thing, set out in Table 3.

The regulations name the Fund’s statutory feeders, proceeds payable under the Betting, Lotteries and Gaming Act, the Income Tax Act’s winnings withholding and the Excise Duty Act, plus appropriations; the Finance Acts of 2025 and 2026 have since rewritten or repealed the taxes behind those very provisions. No public document reconciles what today’s regime actually delivers into the Fund by line: the Parliamentary Budget Office (PBO) projects the excise at KES11.4bn while the Fund targets 24.8, a KES13.4bn gap. The two transactions differ where it matters most: roads sold the future to pay for the past, arrears on works already built, while sports sells the future to build the future. Same instrument, different risk.

2. The Price

Was a middle ground discovered between concessional and commercial money? Judge the pricing at issue against the sovereign’s own auction prints on either side of the close. Table 4 sets them out. The differences are between reported rates, an unpublished-schedule IRR against weighted-average auction yields, not traded spreads, and Section 1’s fork follows the rate here: the auction yields share the market’s NACS shorthand, so if 15.04 is the ordinary quote the raw differences stand, and if the cash’s effective reading governs the comparable figure is 14.51 and every difference narrows by 53 basis points.

The clean comparison is exempt against exempt. The 2040 print, at the notes’ own residual life, gives 76 basis points. The August amortising analogue gives 104. The gross premium around the close therefore runs 76 to 104 basis points on the ordinary reading against tax-exempt paper matched on tenor and structure, roughly 23 to 51 on the effective; the February print, off the tenor match at 11.8 years, sits at 106 and is excluded for tenor, not for size.

Two clocks matter here: residual life, the time left to final maturity, and weighted life, the average time each shilling of principal stays out; amortisation shortens the second without touching the first. Both sides tick on both clocks. The notes amortise by their terms’ 30 redemptions, and every exempt print in Table 4 is a CBK-titled amortised bond with a published tranche structure, so quoted residual life overstates weighted life on both sides. How closely on the notes’ side, nobody outside can say; that depends on the unpublished schedule. On the level model their weighted life runs near ten years, against nine and eleven for the matched prints whose tranche dates are published; if principal is moving late, the notes’ true figure runs longer, and the 15-year prints stay the conservative match on every reading. The shorter prints in Table 4 show yields falling inside the curve.

The taxable prints price a different holder and a different structure. For an exempt holder, who receives their gross yields untaxed, and the two public funds holding 54 per cent of the issue are exempt institutions, the gross differences are the live ones. Against the longer print the difference on the residual clock is 104 basis points. The shorter one is a conventional bullet, its two clocks one and the same, sitting where the notes’ level-model weighted life does: 155. For a taxable holder the comparison moves the other way: the 10 per cent withholding cuts the after-tax return on those June prints, so on the simple approximation the tax-free rate is worth about 16.71 per cent taxable on the ordinary reading, 16.12 on the effective, and the notes’ advantage widens on either; an exact comparison needs each bond’s coupon and price cash flows. A price above the curve on every tax basis and either reading still filled its book with two exempt state funds; the concentration is not explained by tax arithmetic.

Future-flow finance is sold on one promise, cheaper access: collateral is supposed to price the paper below the sovereign’s own curve. This deal carried security over a named levy, an escrow, a reserve and a letter of credit, and still priced above that curve against every matched tax-exempt print, on either reading. So the promise inverted. What the structure bought is not in doubt, the off-book label, exclusion from the official debt stock, a benefit the World Bank’s recount has already begun to erode; how much of the price paid for the label, against the risks investors bore, is the hypothesis the sequel’s print will test. Fees, the reserve and the letter of credit add to the all-in cost, their amounts undisclosed.

The state’s rate is locked for fourteen more years, and the raw differential now stands at about 199-284 basis points against the August 2026 infrastructure bond reopenings of 2035-2042 paper, which cleared between 12.20 and 13.05 per cent. No trading price exists on the restricted segment, so the number is a differential of rates rather than a traded spread, and it bundles limited recourse, illiquidity, novelty, complexity, concentration and fees. The illiquidity is concrete. An IFB qualifies for banks’ statutory liquidity ratios and carries a CBK rediscount window of last resort, per its own prospectus; for these notes, on the restricted segment already described, no such window, qualification or repo use appears on the public record. Buy-to-hold books hold both instruments, so the difference is the instrument’s utility, not the buyer’s intent, and a holder prices it.

The test runs through the sequel, on thresholds set from the fork itself: 50 basis points sits inside what convention and structure alone could explain, 150 exceeds the first print’s premium on any reading. A print within about 50 basis points of the then-sovereign curve into a broad book would compress the hypothesis; a print 150 or more over, absorbed again by the same two public funds, would deepen concern about concentrated demand and weak price discovery, the standing ambiguity a 100.19 per cent subscription does nothing to dispel. Rates fell after the close; that proves nothing about the price on the day. What it creates is a paper gain for holders, the value of collecting 15.04 while the curve sits near 13, borne through the earmark by the Fund’s beneficiaries, and paper it stays unless the restricted segment ever offers an exit. It also raises the obvious question, whether the state can refinance at today’s rates, and the answer sits in call provisions the public record omits: a second missing document beside the schedule.

Table 5 sets Kenya’s four borrowing channels side by side.

The money is the same money: the savings buying this paper are the pool that funds the Treasury’s own auctions, so for at least the anchor half of the book the state changed the price it pays to reach its own savers, not the source. One mismatch rides underneath. The state owes shillings, but part of what the shillings built was imported and paid in dollars, and the roads deal already put a dollar leg inside a shilling-levy structure at pricing nobody outside has seen, powers to inspect it now before Parliament. So the space between concessional money and commercial money is real, but it is not a price; it is a gap between oversight regimes. The observed differential is the visible financing price; the all-in cost stays incomputable and the off-book share of the premium unidentified.

3. What It Is

Three questions hide inside the classification fight, and they resolve differently. In every statistical framework that has ruled, the question is settled: the IMF’s April 2026 assessment of Kenya’s debt statistics states that the road-levy securitisation and similar securitisations of future revenue flows should be recognised as debt liabilities under the international statistical standards, the Government Finance Statistics Manual (GFSM) and the public sector debt statistics guide. The World Bank’s May 2026 assessment already expands the perimeter. It counts 2025 debt at 71.3 per cent of GDP on the wide perimeter against an official 67.3 on the narrow one, with securitised obligations and verified arrears among the additions, a definitional recount, not a policy verdict. Economically, the structure is debt-like: future public revenues service fixed investor claims, and the state stands behind the Fund. Under Kenya’s narrower constitutional and statutory perimeter, the treatment remains contested. The Transport Cabinet Secretary told Parliament the state has no further liabilities once the rights are sold; the National Treasury Cabinet Secretary calls the disagreement an accounting matter. The dispute sits among several items between Kenya and a successor programme since the USD3.6bn arrangement lapsed in April 2025.

Two precedents point the same way for different reasons. The statistical one: Eurostat, in decisions beginning July 2002, ruled that securitisation of future revenues not tied to an existing asset must be recorded as borrowing. Securitisations of fiscal claims are borrowing by convention, because government keeps direct or indirect control of a tax. The decision text goes further in a direction Nairobi should read closely: where government compensates the vehicle beyond its contractual obligations, the entire operation reclassifies as borrowing. Kenya’s reported shortfall safeguards live one step from that rule. The restructuring one: Ghana’s ESLA PLC, created in 2017 to bond energy-levy receivables against sector arrears, was defended in parliament on the same non-sovereign grounds Nairobi argues now. When distress came, its paper was swept into the Domestic Debt Exchange, settled in February 2023 with roughly 77 per cent of its paper tendered against 85 on the sovereign’s own book, restructured alongside the sovereign it legally was not. What remains open is domestic and political: when Kenya’s official statistics, and any new programme, say what the substance already says.

On a household’s books the shape is familiar: a mortgage, the same payment every period until the balance dies. Except a mortgage secures the lender twice, on the house and, in practice, on the income that pays it. This one runs the other way: the receipts are assigned but the stadium is not; the house stays outside the deal. A family that signed these papers would call it debt, whatever the paperwork calls the lender.

4. The Collateral

The literature documented this asset class before Kenya adopted it. World Bank research by Ketkar and Ratha traced future-flow securitisation from the first Telmex transaction in 1987 through more than 200 rated deals worth USD47.3bn by 1999, and analysed its promise for developing-country access; an IMF working paper by Chalk examined securitising public revenue flows as early as 2002. There is no contradiction between that literature and today’s institutional objections: a structure can widen access and still be debt. The mechanism the successful deals relied on is the part Kenya’s version removes. They securitised hard-currency export receivables paid by offshore obligors into offshore accounts, beyond the sovereign’s reach. Kenya’s collateral is a domestic tax, collected and regulated by the state whose behaviour investors are exposed to. The nearest structural cousin is the American tobacco settlement bond, where receipts eroded as regulation did its intended work, stretching maturities and forcing refinancings; the parallel is the collateral’s decay, not the recourse, at least not yet.

Kenya has already demonstrated the mechanism, in dates. Two Finance Acts materially rewrote the betting-tax base within 12 months, while a new Gambling Control Act separately replaced the regulator, and Legal Notice 106 of 2025, in force from 12 June 2025, rewrote the Fund’s own governing instrument, renaming it the Sports and Arts Fund and narrowing its mandate to sports and arts alone. The sequence: excise on stakes at 7.5 per cent in 2021, 12.5 in 2023, 15 in late 2024. Then the Finance Act 2025, assented in the final week of June, abolished the stake excise and the 20 per cent withholding on winnings in favour of 5 per cent on deposits and 5 per cent on withdrawals. Assent came days before the notes closed on 30 June; effect, the day after. Then the Finance Act 2026, effective this July, reinstated the 20 per cent withholding on winnings over the new regulator’s reported objection. The enacted definition is confined to lottery and prize-competition payouts, leaving ordinary sports-betting winnings outside it, narrower than the Bill’s full reversal.

The standing regime after all four rewritings: 5 per cent on each deposit and each withdrawal, 20 per cent withheld from lottery and prize-competition winnings, nothing withheld from sports-betting winnings. Four rewritings of the collateral’s tax law since its 2021 design, two of them rate steps and two base redesigns, the latest already unwinding part of the 2025 design, though not the wallet base the receipts projections rest on. That is the record: collateral rewriting in Kenya is demonstrated practice rather than hypothesis.

The receipts consequence has cut favourably to date, the PBO projecting excise collections more than doubling, from KES5.4bn to KES11.4bn, on the broadened deposit base. The nationally representative FinAccess survey shows participation falling, from 13.9 per cent of adults in 2021 to 11.2 per cent in 2024, so higher collections cannot safely be read as more bettors. The contributions of base redesign, compliance, nominal growth and intensity are not publicly decomposed. Falling participation beside rising collections is also what a per-transaction wallet tax on a concentrating core would produce, fewer bettors transacting more, the reading with the sharpest distributional edge. The wallet design narrows one offshore route, since deposits move through regulated mobile money; it does not close the question. What nothing in the structure covers is the next stroke of the same pen. The protections insure Treasury delay, and no clause reaches statutory redirection. Whether an assignment of receipts survives a future Finance Act is the untested legal heart of every deal in the family.

Rank the family’s collaterals by that logic: the fuel levy broad but hostage to pump-price politics; the rail levy tied to the import cycle; the tourism levy shock-prone; the betting excise most fragile of all, because its erosion is what public-health success looks like. A state that collateralises betting receipts for 15 years hardens its fiscal interest in gambling volumes into a contracted one. Every future stake limit or advertising restriction now carries a debt-service cost, a design property that will outlive its designers. The virtuous cycle is the credit’s headwind: if conversion succeeds and betting fades, the collateral fades with it unless redesign or compliance compensates, and Kenya’s redesigns have so far more than compensated, collections rising through each rewrite. Conversion and the credit are in tension by construction, and the reconciler to date has been the tax code, the least stable element in the whole structure.

5. The Worth-It Test

Two curves run through this decade, and the machine’s fate is the race between them. The first curve is the debt. The observed instalments imply KES6.5bn a year, and modelled service reaches KES12.1-14.2bn once the sequel prices at similar terms, payable for 15 years from the assigned betting receipts. Today the base covers it. Declared receipts lines cover the first deal’s modelled service 2.2 to 3.8 times, depending on basis and fiscal year, and the combined programme 1.75 to 2.05 times on projected receipts, the higher figure resting on the interest-only reading. Both ratios are computed before the Fund’s statutory spending, which the escrow subordinates in practice, and before the sequel trust, whose priority on the same receipts is undisclosed. So sufficiency is not the question. Duration is. The participation base is measurably thinning while collections ride the redesigned taxes, so the asset must build its earning calendar before the receipts themselves turn. That calendar is the second curve, and it matters to the credit through politics rather than cash: renewal in 2028, every further redesign and any Exchequer backstop are political acts, priced cheaply for a full stadium and dearly for an idle one.

The calendar’s one contracted football peak is AFCON, 19 June to 17 July 2027 per CAF’s May 2026 announcement. The tournament closes weeks before the August general election, so delivery incentives run as strong as incentives get, and everything downstream passes through a campaign. Kenya’s designated match venues are Talanta and Kasarani, both in Nairobi, with Nyayo relegated to the training list. In August the state reinstated Eldoret’s Kipchoge Keino Stadium as an alternative match venue, the organising committee’s chairman saying Kenya wants venues in use beyond the tournament. CAF toured it during the eight-day tri-country inspection concluded on 4 September. February’s report found none of Kenya’s proposed stadiums fully met CAF Category 4, itemised Kasarani’s deficit down to pitch reconstruction and 3,000-lux broadcast lighting, set an 80 per cent completion milestone for August, and fixed January 2027 as the deadline for full operational readiness of every match venue and training ground. The September visit returned a positive read on Talanta, the government citing 92 per cent completion, while Kasarani’s works stay under watch.

Kenya has already rehearsed under this examiner. At the African Nations Championship (CHAN) in August 2025, CAF’s reported sanctions on the local federation, fines, a ticket freeze, an attendance cap at Kasarani and a threatened relocation of fixtures, showed enforcement is real. The hosts of the opening match and the final remain CAF’s to announce, reporting presuming the final for Talanta.

The 2029 Worlds belong to the country, not to this credit, because they cannot come to Talanta at all: Talanta is a football and rugby ground with no athletics track, so the championships were never Talanta’s to host, and Kenya’s winning bid stood on Kasarani. So the 1987 stadium captured the decade’s biggest prize while Talanta holds one contracted tournament, and the substitution runs one way, Kasarani able to host Talanta’s codes but never the reverse, which reframes the utilisation question. Box 1 prices the national anchor’s own risk beside it.

Tournament money sits in three pockets that never mix, the organiser’s, the teams’, and the host’s, and the host’s pocket has two sides, outlay and measured gain. Gross-impact projections belong to no pocket at all. Table 6 separates the five.

The organiser’s take is federation income, and the athletes’ take is prize money, paid by results as private income; the winner’s cheque followed the trophy, a fact hosting did not buy. The host’s outlay is what staging costs, and the host’s measured gain is what post-event studies find in direct visitor and event spending, the only pocket a country keeps and the smallest number on the table in every measured edition. Gross economic impact is the issuers’ own projection genus, multipliers and exposure included. That genus is how World Athletics attributes USD586m of impact to a year in which it earned USD69.2m, and how measured gains two orders below the talk attract hundreds of billions of shillings in Kenyan headlines. Organisers expected 1.5 million fans from outside Cote d’Ivoire; the country’s own aviation figures recorded a first-quarter rise of roughly 65,000 passenger movements, attributed to AFCON. Set the measured row against this credit: a full championships hands its host, once, about two to three years of this bond’s service in gross activity. The 2029 edition hands that to the country across town, and the state’s own take is only the tax on that sliver. One measured benefit sits outside all five money types; Section 8 prices it.

Kenya’s own ledger opens on the cost side, and the cost side is the unpublished half. The fee runs host to CAF: KES3.9bn remitted in March 2026 after nearly missing a deadline its partners had met, KES1.6bn for CHAN before it. To that add the KES45.85bn Talanta contract financed by these notes, and a separate KES26.4bn Exchequer earmark announced at the Finance Act 2026 assent, covering the Kasarani and Nyayo works among wider preparation. Add the commuter rail spur whose costs the railway has not disclosed, and an operating model, who runs the stadium and who funds its upkeep, that no public document yet shows, on which Sports Kenya already reports shortfalls. A Principal Secretary said the AFCON budget sat in no current budget; Parliament heard renovations had stalled on unpaid contractors.

Cote d’Ivoire’s precedent prices both columns. It spent about USD1bn hosting alone, one 60,000-seat stadium at USD260m, the benchmark Parliament’s questions about Talanta’s USD344.5m contract now sit beside. And the billion was capitalised rather than consumed, leaving roads, hospitals, hotels and four upgraded grounds the country keeps. Kenya’s consolidated bill exists nowhere in public. Break-even at the sovereign level is therefore not computable by anyone outside government, and the missing documents are cost documents.

Table 7 builds Kenya’s 2027 external inflow bottom-up: matches from the tournament’s format and the CHAN precedent, attendance from the measured Ivorian average, foreign arrivals anchored to the Ivorian aviation record, spend anchored below Kenya’s own average arrival.

The bands are wide where the record is silent, and the verdict survives their whole width. Against annual service of KES6.5-7.6bn on the two bases, USD49-57m at the same rate, the inflow runs from three-tenths of one year’s service at the floor to roughly one year at the ceiling of every assumption simultaneously, a comparison of scale, not coverage. The state’s own take, tax on that inflow at an assumed effective 12-20 per cent, this publication’s assumption, runs USD2m to 10m against the hosting fee of about USD29m paid before a visitor landed. Eldoret’s approval would add a second host city and a few million dollars without moving that range; no allocation outcome changes the verdict. The peaks cannot carry the notes, and nothing in the structure asks them to. No public document maps a shilling from either tournament to these notes; any route is indirect and unmapped, and the analysis runs on the assigned betting receipts, never on event money or multipliers.

That separation is the strongest honest statement of the title’s thesis. The tournament economy keeps its earnings gross, hotels, transport, trade, visibility, while the debt is assigned to the vice base, so no contract connects them. What the tournaments earn the economy still reaches the state as ordinary revenue, easing the budget that appropriates to the Fund and stands behind it on winding-up: separate by assignment, joined through the fisc. A declining vice financing a permanent asset that attracts recurring external revenue is a defensible conversion.

The opportunity side is on the record too: the first AFCON in East Africa in five decades, and CAF’s stated reach of 400 million people. Even the hotels and airlift pass, on the local organising committee’s account of CAF’s own view. The arithmetic above has already priced one peak. And the identifiable outlays beyond the bond, the CHAN and AFCON hosting fees and the Exchequer earmark, already run to about USD240m equivalent before the undisclosed rail and operations. Recurrence is therefore the entire economic case: the identifiable outlays alone equal five to fourteen tournaments of modelled inflow. No tournament anywhere repays the ground built for it; a calendar might. The peaks’ function is to seed that calendar, and CHAN has already banked the first proof.

Beyond the contracted peaks lies a biddable frontier, and to July 2040 it is deep enough to name in tiers. The standing tier is annual and continental: CAF’s club finals already travel, and Zanzibar staged a Confederation Cup final leg in 2025. The contest tier is dated and priced: rugby’s new Nations Championship finals opened a neutral-venue market at a reported GBP800m over eight years, London 2026 then Doha 2028, with later editions returning to the market if targets slip. The one-off genre is live, a Test of the Washington 2018 kind. The ambition tier is what the estate was sized for: a solo Kenyan AFCON in the quadrennial era that begins after 2028, each edition scarcer and richer for it, and an African Commonwealth Games in 2034 still unawarded. Each prospect divides by the same one-way line, football and sevens to Talanta, track and cross country to Kasarani; the athletics frontier fills the other stadium with every prize the country wins.

Every tier above the contracts is a bid, and bids are won on the readiness record the examiner is writing now; 2027 is the audition for the whole calendar. Peaks justify scale; the payments are made on the floor, the league finals, concerts and gatherings of a growing city. And the floor now has more claimants than the order book assumed. The programme has two physics, the flagship carrying the order book while the 33 county facilities carry the pathway, and averaging them flatters whichever is failing. Should CAF approve Eldoret, the state exits the decade with three large grounds, two of them 25 kilometres apart sharing one city’s bookings, the third in a city whose sporting identity is Kasarani’s code. The only Kenyan evidence on that floor is Kasarani itself, built for the 1987 All-Africa Games, carrying the national stock alone for four decades, its ordinary calendar undocumented; Cape Town’s Green Point and much of Brazil’s 2014 stock name the genre this build must escape. The 60,000 seats on Ngong Road will be judged on what fills them between the fortnights the world visits. And the judging is a race: that calendar must establish before the collateral turns, because the bridge between the two is the tax code.

6. The Demographic Ledger

The Saturdays and the payments are serviced by the same constrained flow of young Kenyans’ money, which is where the ledger turns demographic. Who is in this transaction? Start with prevalence. The nationally representative FinAccess survey puts active betting participation at 11.2 per cent of adults in 2024, down from 13.9 in 2021. That is the participation measure; the survey’s separate betting-as-income series printed its own 11.2 back in 2021, and the two are distinct. Mobile-panel surveys of the connected young run far higher, an upper bound for the digital cohort rather than the nation. Young people are overrepresented on several legs of the structure, to different degrees on each. The same survey puts male participation at 18.4 per cent against 4.4 for women, highest in the 18-25 and 26-35 brackets. They are overrepresented among bettors, among the Fund statute’s intended beneficiaries by its own text, in the construction labour of a rebounding sector, and, through the NSSF’s widening enrolment, among eventual claimants on the creditor side. Today’s payout accrues mostly to established balance sheets, over half of it to two public retirement funds.

Kenya’s population skews young by any bracket: roughly three in ten Kenyans are aged 18-34, over half of adults, and about three in four are 35 or under. Something near a million people enter the labour market yearly on the standard estimate, into youth unemployment near 15 per cent on the strict ILO definition and roughly double that on broader measures of those seeking work.

Where the education-to-employment channel is congested, alternative pathways gain value, and sport is one of the few where Kenya exports at the global frontier. The verdict question is precise, and it is still a hypothesis awaiting evidence. The channel is virtuous if it converts the revenue base into the productive base. It is regressive if two hundred shilling deposits, drawn disproportionately from young men, fund a 15 per cent tax-free return to established pension balance sheets against a monument. Conversion can be measured: county facilities completed and used, academies and athletes coming through, event-days and paid attendance at the stadium, jobs in the sector. The overlap between the transaction’s legs is not measured, and the incidence is undemonstrated either way. The strongest defence of the whole practice belongs here at full strength. The window is open now. Saving receipts for 15 years first would serve a cohort already aged past the pathway, so time-shifting capital to meet the cohort while it is young can rationally justify borrowing. Its limit: the case licenses borrowing at the sovereign’s own rate, on the books. Nothing in it requires the differential, or the opacity of structures Parliament cannot inspect.

7. What Travels

Nigeria, Ghana, Uganda and Tanzania all tax betting without a conversion channel, and Kenya’s experiment is the one their treasuries will study. Four conditions on the structure and one on the project decide whether it should travel. Count the obligations from day one; the statistical and restructuring precedents have both ruled on models that hide. Ring-fence in law that survives the Finance Act cycle, the live threat Kenya’s own record proves. Build pension demand that is market-made. African retirement capital will buy 15-year local paper, an encouraging discovery only if the buying was chosen, and replication needs a broad book rather than two state-adjacent anchors: an auctioned or widely syndicated book, priced against a published curve, is what chosen looks like. Choose collateral for political durability; a levy whose erosion is a policy success is the weakest class. And contract the order book before pouring concrete. What should not travel: dollar legs inside local-levy structures, sunset clauses that park an undisclosed statutory residual on the Treasury, and single-event business cases. Whether Kenya’s own version clears these conditions is what the verdict now assembles.

8. The Verdict, in Instalments

Count it as debt in every ledger beyond Nairobi’s statute book; the domestic classification is a political timetable rather than an analytical question. The opening’s two outcomes are now the two live questions, and both stay open here. Arbitrage is the premium question: secured paper priced above the sovereign’s own unsecured tax-exempt curve on either reading, the promise of cheaper access inverted; whether the premium was earned, against the counterfactual of a mid-2025 infrastructure bond at the sovereign’s own curve with undisclosed fees and enhancement costs added, is what the sequel facility’s print and book will inform. Architecture is the keep question: the one contracted peak is priced and cannot carry it, four months to a year of service, once; the keep is earned on conversion, the county programme, the operating model, and the ordinary calendar.

The observables are dated. First, the terms on which the Sports Fund is renewed ahead of its 2028 sunset, August on the memorandum, October on the regulations’ commencement and the ten-year rule: non-renewal would default paper held by the state’s own pension funds, so renewal is the likely path. The discriminating question is whether the earmark and the assignment survive intact, since June 2025 rewrote both the tax base and the Fund’s own name and mandate; redesign reaches the instrument itself. Then the KES38.74bn print and who buys it; the January readiness deadline; receipts against the reported payments, and the year they first turn, the moment redesign stops compensating for a thinning participation base, visible only if the state publishes the decomposition. The rest of the list: priority between the two trusts once the sequel’s documents surface, secondary prints if the restricted segment ever produces one, and who operates and maintains what has been built.

One benefit is real and measured, and it sits in no ledger above. The wellbeing literature finds hosting is the channel that moves a population. Across twelve European countries, Kavetsos and Szymanski measured a large and significant lift in reported life satisfaction from hosting football tournaments, while the effect of winning alone was statistically indistinguishable from zero. Dolan and co-authors found London 2012 lifted Londoners’ happiness during the Games, peaking at the ceremonies, gone within the year, and worth its cost only if the rest of the country shared a modest part of the feeling. Kenya has consumed the winning kind for four decades and is buying the hosting kind for the first time. A first African Worlds and a first East African AFCON in five decades present every condition the literature ties to its largest effects: hosting rather than winning, novelty, national reach. Pride of that order is a legitimate public purchase. The question that remains is the price and the financing, because the lift fades by the following year and the schedule does not.

Who bears a failure is already filed. The Fund’s statutory beneficiaries absorb the first fiscal adjustment; Treasury payment of a winding-up deficit is a statutory duty conditional on National Assembly approval, and neither the approval nor direct noteholder recourse is automatic. That is the public, as beneficiary and pensioner, and potentially as taxpayer, on both sides of a trade currently paying its pension side handsomely, on paper.

The young man’s two hundred shillings has a long itinerary. The stadium his deposits are building opens for a continental tournament in June 2027; the championships that crowned award week play across town in 2029; the last payment his successors will fund falls due in July 2040. The verdict arrives in instalments between those dates, and the first two are already scheduled: CAF’s readiness deadline in January 2027, and, likely sooner, the print on the next facility. The race inside those instalments is simple: the ordinary Saturdays must arrive before the betting base turns. Whether the machine converts vice into virtue rides the same calendar, county grounds and operating model beside the prints; like the debt, the cycle will be counted, not believed.

Sources

AFP, AFCON 2023 preview reporting on expected foreign attendance (January 2024); Athletics Integrity Unit and WADA compliance record, as reported (Monaco and Nairobi, 2022-2026); Bowmans and McKay Advocates, Finance Act 2025 client alerts (Nairobi, July 2025); Business Daily, reporting on the Linzi FinCo 003 offer terms, the Fund sunset provisions in Kenya Risks Asset Seizures over the Talanta Bond, current-year payments, the roads bond target, corporate bond market and pension holdings from RBA filings (Nairobi, 2025-2026); Cabinet Office, statement on road sector pending bills (Nairobi, December 2025); CAF, AFCON 2027 kick-off and final date announcement (May 2026), inspection releases (September 2026), the KSh3.9bn 2027 hosting-rights fee as reported (March 2026), financial statements to the 47th General Assembly and President Motsepe’s AFCON 2023 profit and prize statements, AFCON 2023 host-economy markers as reported (2024-2025), CHAN 2025 fixtures and venue allocation (2025), AFCON 2027 venue inspection reports as reported (February and September 2026), the Zanzibar Confederation Cup final staging (2025), the post-2028 quadrennial cycle announcement as reported (December 2025), the Morocco 2025 profit projection and prize structure as reported (2025-2026), and the commercial director’s AFCON 2027 combined-impact projection as reported (May 2026); Central Bank of Kenya, 2024 FinAccess Household Survey (Nairobi, 2024); Central Bank of Kenya, Treasury and infrastructure bond auction results, February to August 2025 and August 2026, as reported (Nairobi, 2025-2026); Chalk, Nigel, The Potential Role for Securitizing Public Sector Revenue Flows, IMF Working Paper 02/106 (Washington, 2002); Controller of Budget, county pending-bills reports (Nairobi, 2026); Daily Nation, AFCON 2027 venue and inspection reporting (Nairobi, 2026); Dolan, Paul, Georgios Kavetsos, Christian Krekel, Dimitris Mavridis, Robert Metcalfe, Claudia Senik, Stefan Szymanski and Nicolas Ziebarth, Quantifying the Intangible Impact of the Olympics Using Subjective Well-Being Data, Journal of Public Economics 177 (2019); Eurostat, decisions on securitisation operations undertaken by general government (Luxembourg, 2002 and 2007); Finance Acts 2025 and 2026 as recorded by Kenya Law, with the 2026 winnings provisions as reported; Gambling Regulatory Authority and Kenya Revenue Authority, gambling-tax figures presented to the National Assembly, as reported (Nairobi, May 2026); GeoPoll, Betting in Africa reports (2022-2025); Ghana Ministry of Finance and Domestic Debt Exchange records on E.S.L.A. PLC, including the 13 September 2023 reopening release with instrument-level tender rates, as reported (Accra, 2017-2023); IMF, Kenya: Data Quality Assessment for Public Sector Debt Statistics, High-Level Summary Technical Assistance Report 2026/020 (Washington, April 2026); Ivorian civil aviation authorities, first-quarter 2024 passenger figures as reported (Abidjan, 2024); Kavetsos, Georgios and Stefan Szymanski, National Well-Being and International Sports Events, Journal of Economic Psychology 31 (2010); Kenya Railways, Nairobi commuter-rail works for AFCON as reported (January 2026); Kenya National Bureau of Statistics, Economic Survey 2026 and census analytical reports (Nairobi); Kenyan Wallstreet, Linzi 003 listing and amortisation analyses and CBK auction results (Nairobi, 2025-2026); Ketkar, Suhas and Dilip Ratha, Development Financing During a Crisis, World Bank Policy Research Working Paper 2582 (2001) and Innovative Financing for Development (2009); Lawton, Matt, The Times (London, 15 September 2026); Ministry of Sports Principal Secretary, venue readiness statements as reported (Nairobi, September 2026); Ministry of Tourism, 2024 tourism earnings and arrivals as reported (Nairobi, 2025); Mwango Capital, market reporting on the February 2025 infrastructure bond reopening (Nairobi, 2025); National Assembly and Senate records on the Talanta contract and securitisation oversight (Nairobi, March and May 2026); National Treasury, 2026 Budget Policy Statement (Nairobi, February 2026), Annual Borrowing Plan FY2025/26 and Budget Statement FY2026/27, as reported; Parliamentary Budget Office, Budget Watch 2025 (Nairobi); Pergamon Investment Bank, Fixed Income Pre-Auction Notes carrying CBK auction results tables (Nairobi, 2025-2026); President’s Finance Act 2026 assent statement on the sports-infrastructure earmark, as reported (June 2026); Public Finance Management Act, Cap 412A, section 24(9), and the Public Finance Management (Sports, Arts and Social Development Fund) Regulations, Legal Notice 194 of 2018, as amended by Legal Notice 106 of 2025 (Nairobi); Six Nations Rugby, SANZAAR and World Rugby, Nations Championship structure and finals hosting, London 2026 and Doha 2028, as reported (2024-2026); The Star, reporting on projected 2029 Worlds economic impact (Nairobi, September 2026); Wales v South Africa Washington DC Test, as reported (2018); World Athletics, 2025 Annual Report and Accounts (September 2026), Tokyo 2025 prize schedule, attendance, staging cost and prior host-city gain estimates as reported (2017-2026), and award coverage including Reuters, AP and BBC reporting (September 2026); World Bank, Kenya debt assessment (Washington, May 2026).

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: The Second 43.

Exactly a year ago today, I wrote to you with one eye closed. It was a Saturday morning in Nairobi, Wellington was under lights, and the scoreboard read 43-10. The heaviest defeat in All Blacks history, and it came at home, after a second half that finished 36-0. I called it a diagnosis and asked New Zealand Rugby for a Tuesday. A scrum that resolves instead of resets. A school built from the hardest nights. Work beginning in February, so the jersey could meet them in July.

The second 43 arrived overnight. Past midnight in Nairobi, in a borrowed NFL stadium in Baltimore, South Africa closed out the All Blacks’ first full tour of South Africa in thirty years. It finished 43-28, the series won 3-1 after losing the opener. Two Septembers, two 43s, one direction. On the scoreboard, the losing margin narrowed by 18 points year on year. Watching it live said otherwise. The match stayed alive into its final ten minutes, and every time New Zealand fought back within reach, South Africa found another acceleration. Beauden Barrett’s last-minute try narrowed a margin that had touched 22. A year of Tuesdays has passed, and the report card came back in green and gold.

Ardie Savea spent this series as a captain Springbok supporters openly adore. He was injured in Soweto after scoring twice and was gone by Baltimore, and the affection did not pause for his absence. They cheer him. They call him their favourite non-Bok. Richie McCaw spent a career being called a cheat in the Republic, where the chorus held that referees let him get away with murder at the breakdown. Dan Carter was respected the way you respect a storm. Some of the warmth for Savea is simply the man. The way he plays and carries himself is legible to South African rugby culture in a way few All Blacks have ever been. The Cape has always held its pockets of genuine All Blacks support besides. But I do not think the character changed. I think the tax on admiring it did. Springbok supporters could never afford to see McCaw whole, because he was the edge of their defeat every Saturday. They can see Savea whole because his brilliance no longer threatens their hold on the rivalry: South Africa have won 8 of the last 10 tests between these teams. The affection is not the proof; the results are. I read it as a tell. Fear distorts perception, and dominance is what removed the distortion.

Baltimore had its contestable moments, as every test in this rivalry does. A loose ball spilled from an All Black who many in black swear was tripped, waved on, and scooped by Kurt-Lee Arendse inside his own half for a fifty-metre score. Cam Roigard was taken out after kicking through, a certain penalty until the television official went back a phase and found him offside first. A stoppage of more than eight minutes in which the match officials’ communications failed. Reasonable people can argue each call. What interested me was who was doing the arguing. Through the long decades of All Blacks supremacy, relitigating the referee was a Springbok supporter’s pastime, and we in black found it faintly amusing. Overnight the roles ran the other way, on my own timeline: All Blacks fans working frame by frame, Bok fans laughing. Sentiment trails results as an after-image, and the affection and the grievance now point in the same direction as the scoreboard.

The coaching box holds another tell. Scott Robertson won 20 of his 27 tests as All Blacks coach, 74 per cent, and went 1-3 against South Africa. He departed in January after an internal review, with friction between senior players and team staff reported through the preceding month. Dave Rennie replaced him in March carrying a career record of beating South Africans, including a 3-1 record against these very Springboks as Wallabies coach. His first series in black finished 1-3. The same man, against the same opponent, four years apart: 3-1 in gold, 1-3 in black. This is no controlled experiment; squads, staff and settings all differ. But it is the sharpest hint yet against the magic-wand theory of coaching, because if the coach were the binding variable, the advantage should have travelled with the man, and it inverted. It is the pattern I wrote about a year ago. My All Blacks have become my Manchester United, I said then. Post-Ferguson United is precisely the case of a club that kept changing managers while the deeper problems of authority and direction survived every rotation. Two consecutive All Blacks coaches now stand 1-3 against South Africa, and one has already lost the job. My reading, and I mark it as a reading, is that player dissatisfaction materially shaped the review and its outcome. Reporting at the time framed it as player power; the union’s public line was otherwise. If that reading is right, the next man should study his predecessor’s authority even more closely than his game plans.

The year kept asking me one question: what is dominance actually made of? Answering it meant no longer trusting my memory of eras, and counting instead. I rated every test played among the ten nations of the Six Nations and the Rugby Championship, from 1987 to Baltimore, on a continuous rating of my own. Every ranking number in this piece comes from that model, not from World Rugby’s table, and the two run on different scales. At every checkpoint where they can be compared, the six tournament eves since official rankings began, the November 2024 handover, and the two summit interruptions since, they agree at all but one. That one, 2019, is marked in Table 2. What the counting found is that dominance is one architecture read on two gauges, and that winning World Cups is a different architecture altogether. The first gauge is the summit: how far above the entire field a team sits, and for how long. The second I will call the form line: how well a team plays against top opposition on its own terms, regardless of who else is great at the time. Conversion, the machine that wins the three weeks of a tournament, is not a third gauge. It is the other architecture, and it keeps a clock of its own.

On the summit gauge, one rule was set before counting: reigns begin and end on match days, wherever the results put them, and never by the tournament calendar. Sometimes the breaking result is itself a World Cup match, and the match still governs; the 1987 final opens the second reign on this table and the 1991 semi-final closes it. What the rule forbids is cutting eras at tournaments by convention, because dominance does not respect the four-year calendar. The model keeps 21 reigns at the summit since 1987, counting the Wallaby reign already under way when the record opens, and many of them are caretaker spells lasting weeks. All 21, ranked by depth, which is average gap multiplied by duration in years, computed on unrounded figures. The gap is always the rating distance to the second-placed side. The win rate is each side’s top-tier record across the reign’s own tests. The ranking runs on the summit gauge alone. The second-last column sits outside it and marks what conversion did inside each reign. It distinguishes titles that opened a reign from tournaments entered while holding the summit:

The deepest reign starts and ends on match days, the day the All Blacks overtook the World Cup-holding Springboks and the day of a 47-26 defeat in Perth, neither endpoint set by a tournament result. The form line says something starker still: from mid-2002 to late 2020, eighteen and a half unbroken years, New Zealand’s rolling win rate against top-tier opposition never once dropped below 7 in 10. Third on the table sits the first Henry reign. My model dates it from March 2004 to August 2009, ending in the season John Smit’s Springboks beat them three times. Henry was still in the chair when the next reign began two months later. Within his first reign I isolate one window, 2005 to the 2007 quarter-final, cut at the tournament because the tournament is the question being asked of it; the reign’s boundaries stay match-driven around it.

A British and Irish Lions series swept 3-0. Three consecutive Tri-Nations. The first All Blacks Grand Slam of the professional era. Across the window they won 87 per cent of their top-tier tests at an average margin of 18 points, a core hotter even than the reign around it. On the tournament’s eve, the model had them 306 points clear of the entire field, the most daylight any side has carried into a World Cup in the forty-year record. A month later they were out, beaten by France in a quarter-final, the earliest exit in their history. Of the three deepest reigns ever measured, theirs is the only one that never lifted the trophy: the deepest holds the 2011 and 2015 titles, and the second was launched by the final of 1987.

The form line exists because the summit gauge is blind in one direction: a great side that coexists with a greater one is invisible to it. Clive Woodward’s England ran at 78 per cent across five years, average margin 16. They were blocked from the summit for most of that run because the All Blacks and Wallabies stood on it. The model crowns them only in the spring of 2003. Read on that line instead, and the full canon comes back. The Wallabies of the early nineties and of the turn of the century, Woodward’s England whole, the Springboks of Mallett and of Smit, France and Ireland in this decade. Nick Mallett’s side is the one time South Africa tried the conventional order, building a dominance era first, 17 straight wins, and it remains their one great era without a title. Every South African title came the other way around.

On the summit gauge, the current Springbok reign’s average distance above the field, 132 points, is larger than any non-New Zealand reign in the record. On the form line, their cumulative record since late August 2022 runs at 80 per cent, three points shy of the Henry-era All Blacks across a full reign and ahead of peak Woodward. The line crossed the model’s 75 per cent threshold on 22 June 2024 and has not dipped below it since. Two clocks: the form arrived in the middle of 2024, and the summit followed that November. If the reign reads short against everything it measures, Table 1 says why. Until that November weekend the summit was Ireland’s, row seven, held for two and a half years while the world champions chased it on the rating. The crown never moved in those years; the summit still had to be taken. The gap after Baltimore stands at 163 points, wider than any reign outside New Zealand has ever averaged. Four reigns carry a higher average than this one, all of them New Zealand’s; the Wallaby reign that sits above it on the table ranks there on duration, not altitude. The deepest of the four needed a decade to accumulate what this side has banked in under two years. The reign is still open; its averages can still rise or fall. On both gauges, the most dominant team in the world today is South Africa, by a distance.

But dominance, on either gauge, is not conversion. South Africa has won four World Cups, more than anyone, and has never once entered one as the top-rated side:

Official rankings begin in 2003; before that, only the model speaks. Where both exist they name the same leader at every eve except 2019, where the model’s 16-point England edge is noise and the official table says Ireland; on either reading, not South Africa. Four titles, and on all 4 eves somebody else was the dominant team on the planet, and every one of those 4 leaders fell. Across all 10 tournaments the eve leader converted 4 times, 1999, 2003, 2011 and 2015, and of the 3 sides that arrived commanding, more than 150 points clear, 1987, 2007 and 2015, exactly 1 converted. A champion must win three knockout matches; grant a genuinely great side an 85 per cent chance in each, against top-tier knockout opposition, and it clears all three about 6 times in 10. The model is illustrative and the matches are not coin flips, but it makes a team the favourite without making its failure surprising. Dominance is advantage, not insurance.

The familiar ambition, the one every side on the reigns table pursued, is to build the machine that wins the years and pray it converts when the three weeks arrive. Under the Erasmus programme the outcomes arrived in the opposite order. Across 2019 and 2023 they played 6 World Cup knockout matches and won all 6, 4 of them by 3 points or fewer. The last 3 came by a single point each, against the hosts, the English and the old enemy. They did it while losing pool games at both tournaments. Between those titles, the pandemic took twenty months of tests from them, and the results they did play were uneven in a telling way. A British and Irish Lions series won in the middle of the valley, the weeks machine firing while the years machine idled. Then back-to-back defeats to Australia a month later, 5 more losses in 2022, and Ireland holding the summit at 78 per cent of their own. Whether the order was designed or discovered, only Erasmus knows; the record only certifies the sequence. Conversion first. The years machine, second.

Any Springbok supporter worth their replica jersey dates this era from Yokohama, November 2019. Seven years and two titles, never surrendered. On their instrument, they are right. Call it the champion’s clock: it ticks only at World Cups, because the crown can only change hands there. On that clock South Africa’s tenure has run unbroken since then and will arrive in Australia at eight years. That matches the longest championship tenure any side has ever carried into a tournament, the All Blacks’ own 2011 to 2019. It is silent between tournaments by design, roughly 47 of every 48 months, and it measures the one possession those months cannot move. The clock could not see 2021 or 2022, just as it could not see the 2005 to 2007 All Blacks. The clock is not in dispute; the other two instruments cover the years it ignores. November 2024 is when the argument about the era ended. That is when the last of the three clocks, the summit, came into phase with the champion’s clock, the form line having joined that June. South Africa have aligned all three only once before, under Mallett across 1998 and 1999, and that alignment dissolved before the tournament arrived, where Australia took the crown as well. This one is their first with back-to-back titles already banked, and the first still intact on the road to a World Cup rather than spent ahead of one. Their four titles were all won while the other clocks read someone else’s name; since late 2024, all three read the same one. The summit wobbled twice, at Ellis Park both times. The line has held since June 2024. The champion’s clock has not moved since 2019. In a programme built conversion-first, that is the one you would expect to be the stillest.

And the second machine is now unmistakably built, complete with the signature the first one always had. They win when it counts. They still tune out between, in miniature. A 16-point home defeat to Australia at Ellis Park in August 2025, met seven days later in Cape Town. A 67-30 demolition of Argentina, nearly squandered in a two-point title-decider escape seven days later, the flicker inside a single fortnight. A 16-33 first test at Ellis Park three weeks ago, followed by three straight. The summit line itself shows the same heartbeat. South Africa took No 1 over the weekend of 8 to 10 November 2024, on the Friday, without playing, when Ireland fell in Dublin. They confirmed it on Scotland two days later. They have surrendered it exactly twice since. Four weeks after the 2025 defeat at Ellis Park, taken back by the first 43 in Wellington. One week after the 2026 defeat at the same ground, taken back in Cape Town. On the field, each defeat was avenged at the very next test, seven days on; the summit simply took longer to follow the first time, and the official table flickered at the same two moments. The gap has widened through all of it. 97 over Ireland in January 2025. 144 after a July in which England, Scotland and Wales came to the Republic and conceded 130 points for 49, Wales departing 43-0. And 163 over New Zealand after Baltimore. They still concede episodes. They no longer concede seasons. And all of it is one Erasmus-built system: coaching, selection, depth and institutional authority running as one programme. One regime, through his own head coaching, the Nienaber years with Erasmus as director of rugby, and his return. Erasmus-led continuity since 2018 is no longer a coaching variable but institutional capacity, which is exactly what New Zealand rotated instead of built.

Three sides have entered a World Cup as both the reigning champion and the top-rated team in the world. The All Blacks of 1991, who fell in a semi-final. The Wallabies of 1995, who fell in a quarter-final. The All Blacks of 2015, who converted. 1 from 3, the same rate the commanding favourites post. South Africa in 2027 would be the fourth, and the first of them to arrive as double defending champion.

A year ago I posed questions and issued prescriptions. Table 3 answers each in turn:

The leadership drawer I parked last September is open. The appointment was made, and what it revealed sits in the affection above. What it has not yet revealed is where authority actually lives between players, coach and union, and this year made that question larger, not smaller. The scrum: in Soweto the penalty try came late in the first half and the maul added another through Marx, in a five-point thriller. In Baltimore the decisive early second-half break was launched from a scrum penalty and the maul that followed, with another scrum the platform for the next score. New Zealand’s set piece did not lose every contest this series. It lost the contests that decided the matches, and that has been true three seasons running: the maul that powered the Ellis Park comeback in 2024, the Wellington night that named the sickness in 2025, and now Soweto and Baltimore. The school: I can find no public evidence that anything resembling the curriculum was built, and what happened instead was cheaper. Changing a coach requires one decision; a curriculum requires staff, habits and years, and there is no public sign anyone priced the difference. Add Baltimore to the wall. Eight postcards now, Townsville, Mbombela, Twickenham, Paris, Johannesburg, Cape Town, Wellington, Baltimore, and still no timetable on the door.

The grindstone thesis received the strongest stress test it has yet been given, and New Zealand’s year was not empty. They swept their July, 3 from 3, Ireland put away at Eden Park. So did South Africa, by more, and the tour was the tiebreak. The tour gave the All Blacks six weeks inside South Africa’s world, 4 franchise matches and 3 tests in the Republic before the Baltimore finale, the old immersion no calendar had offered in thirty years. They won all 4 franchise games, one of them an escape from 35-17 down against the Lions, and lost the test series. Sharper still, they won the test played earliest in the immersion and lost the 3 that followed; more exposure did not mean better results. A sceptic will answer that South Africa habitually start slow after a break, Ellis Park 2025 says as much, and the clause is fair. But the narrow finding stands: six weeks of immersion could not substitute for the operating year. My reading of the deficit is unchanged, the weekly diet and the Februaries that were never used, and one August cannot buy them back.

The morning after Wellington my short-term prescription was a slower, structured game, boring by design. The series’ only All Blacks win came from the opposite. Rennie’s expansive opening night at Ellis Park, by his own numbers, converted at 3.7 points per entry into the Springbok 22 while holding the world champions under one point per entry. It was the best attacking answer New Zealand have produced in this fixture in years, and it was not my answer. Read the whole night, though, and it was not pure expansion either. South Africa’s power game controlled much of the first hour, and the win was built on absorption plus a lethal strike rate, five tries to one from limited entries. And then read the three defeats, each decided precisely where my prescription pointed, Kolbe’s boot and the lineout steals in Cape Town, the scrum in Soweto, the set-piece launches in Baltimore. So the tactical prescription lost the argument about how to win, and the structural diagnosis won the argument about where New Zealand remain vulnerable. In this series, expansion took the opener and did not survive South Africa’s adjustments. That settles nothing about 2027; it does move the burden of proof.

The coda to those essays argued that structure and rhythm outlast slogans and star names, and that unglamorous Tuesdays decide September outcomes. New Zealand Rugby’s year inverted it: the star names outlasted the structure, no operating calendar was ever made public, and September arrived on schedule anyway.

A double world champion sits sixth on Table 1 because that table reads the summit clock alone, twenty-two months old for them, while their greatness kept the champion’s clock. If the current reign holds, they arrive in 2027 holding what only three sides in the record have ever carried in: the summit and the proven conversion machine at once. Their gap after Baltimore, 163, stands beside 2007 and 2015 among the largest ever taken toward a World Cup. Those two precedents should be studied in that order, the side that fell in a quarter-final first. The eve leader’s record is 4 from 10; the commanding favourites’ record is 1 from 3. The draw will still hand South Africa three knockout matches against the best sides alive. Their two machines alter those odds. They do not repeal them. One warning is the fate of the last such alignment before theirs: the All Blacks carried all three clocks from 2015 until Perth, six weeks before the 2019 eve, and fell in a semi-final. Thirteen months of tests stand between Baltimore and the eve. The other is written in their own signature. Their defeats stay episodes because an answer always comes within a week, and a knockout tournament is the single arena that cancels the answer week. Three of the 6 who fell died in the quarter-final. The thing this team concedes is precisely the thing the format refuses to forgive. If they lift the trophy, they will have converted from the front after a history of converting from everywhere else. If they fall, the reign joins the clearest demonstrations in forty years that three knockout weeks guarantee nobody anything, and nothing about the fall would erase what the programme already owns.

As for my team. New Zealand enter the cycle’s final year without the summit or the form line. They carry a set piece still unanswered in the moments that decide tests, an authority structure I no longer recognise, and a coach who has just watched what happened to the last one. I chose this team in 1995 for a standard, and I am not leaving. But I wrote a year ago that the prescription was work that needs no permission, and I have stopped asking for Tuesday. The men who run New Zealand rugby know the timetable. February comes around every year. The next one is the last one before Australia.

One eye is still closed. The other one is on the counting.

Data and method note: every rating, gap, reign and win rate above comes from the author’s own Elo-family model, built on all 2,886 tests played among the ten Six Nations and Rugby Championship nations from 1871 to 12 September 2026, with the analysis window opening in 1987. The scale is the model’s own and is not comparable to World Rugby’s ranking points. Win rates are top-tier rates: fixtures against nations outside the ten, and British and Irish Lions series, are excluded in every era equally. Reign boundaries are match-driven. A reign absorbs a single interrupting spell shorter than 90 days only when the same side stands on both ends of it, a tolerance applied to every reign equally: the current South African reign absorbs the two spells named in the text, the 2009 South African spell of 91 days stays unmerged by one day, and the multi-sided 2019 interregnum between the two New Zealand reigns is never merged because no single spell in it is flanked by the same side. The form line is a trailing 20-test win rate, and South Africa’s crossed the 75 per cent threshold on 22 June 2024. Both detections run identically for all ten nations, and the full summit record holds 21 reigns. South African eras register at the summit in 1998 to 1999, for three months in 2009, and for a thin pandemic-era year from February 2021 held at a 34-point average gap without the form line ever joining it; on the form line they register across 1996 to 2000, 2007 to 2009, 2011 to 2014 and 2022 to date. At the six World Cup eves since official rankings began, the model and the official table agree on the No 1 side at 5 of 6; the sole divergence, 2019, is shown in the table. The headline findings, including South Africa’s absence from every tournament-eve summit and both 2025 and 2026 summit interruptions, are stable across the tested parameter ranges; precise figures such as 306 and 163 are baseline values. Scoring-system changes across eras, including the 1992 try revaluation, affect raw margins and are noted accordingly. The reign ranking is metric-robust at the top: gap alone and gap multiplied by duration select the same four reigns, all New Zealand’s, though in a different internal order, gap alone placing 1987 to 1991 first; the only ordering that decides anything in the text is fifth against sixth, where average gap favours South Africa and depth favours the longer Wallaby run, and both readings are given.

AI-illustration: The ledger opens in the first hour.

A note before we start. This discusses the third season of House of the Dragon, which ended this month, and then the ending recorded in the written history behind it. That history is George R. R. Martin's Fire & Blood and, for the dynasty's later years, The World of Ice & Fire, which Martin wrote with Elio M. García Jr and Linda Antonsson. The series departs from that history considerably, and the parts I lean on hardest are among the departures. The prophecy on the dagger, the misheard deathbed, Alicent's offer of the city, the killing of the High Septon, Corlys held at Tumbleton and the parley there are not in the published account. I have watched the series and not read the books, so the closing material comes from published accounts of a chronicle that disagrees with itself. It is the record's ending rather than the series', and the series has a season still to run. In passing, it glances forward to Game of Thrones and to how that series ends; the books have not reached that far.

I watched the third season of House of the Dragon earlier this month, eight episodes across three nights, and came out of it more irritated than entertained. The irritation stayed with me for a fortnight, and it took most of that time to work out what it was about, because it had very little to do with the quality of the storytelling.

The first explanation I reached for was the characters. Most of the people with a claim to power in that story spend the season securing whatever position they are standing in. I kept waiting for one of them to take a decision that would still be doing useful work years later.

Then I got to Viserys, and I understood what had been bothering me. It is not that he does nothing. It is what he is willing to pay for, and what he is not.

The ledger opens in the first hour of the series. Viserys’s queen, Aemma, is in a difficult labour, the child is breech, and the maester tells the king he must sacrifice one or lose both. Viserys gives permission. Aemma is never told what is about to happen; he says only that they are going to take the babe out. Midwives pin her to the bed, she sees the blade, and she screams. The child is a boy and he dies within hours.

The maester had told him she would very likely die either way. That narrowed the choice. It did not explain why the person being cut open was never allowed to hear it, or answer it.

So we know at the outset what this man is prepared to decide on someone else’s account, and how fast. He was given a choice and made it at once. The woman who would pay for it was never asked.

The problem in front of him was never a shortage of instruments. In the year 101 after the Conquest, a Great Council sat at Harrenhal to settle a disputed succession. It weighed the claimants and gave the throne to Viserys over his cousin Rhaenys, who in the series’ telling stood in the senior line. The lords chose the man, and the choosing is what made Viserys king.

He knew what the instrument was. He had been its beneficiary.

When his own turn came, he did not use it. He named his daughter Rhaenyra his heir and had the lords of the realm swear to her in his hall. Afterwards he told her the family’s oldest secret, Aegon the Conqueror’s dream, and charged her to carry it. The secret was not held in his head alone. It is inscribed in High Valyrian on a Valyrian steel dagger, legible when the blade is heated, and later he warms it and has Rhaenyra read it aloud. Within a few years he married again, to Alicent Hightower, the daughter of his Hand, Otto Hightower. She gave him sons. The eldest was named Aegon, after the Conqueror. The king now had a son. Still, Viserys did not change the succession. Rhaenyra stayed his heir, and the oaths stood as sworn.

The naming, the oaths and the dagger each declared his answer. None of them required him to defeat the rule standing against it, that a son comes before a daughter, and none of them cost him anything, which is why they were insufficient.

The test comes on the night Viserys dies. His council meets in secret before dawn, and the plan on the table is to set Rhaenyra aside and crown Aegon in her place. That night and in the day that follows, the dagger and the oaths are both put to the test.

The dagger fails because it is not a rule. Anyone who knows to put the blade in a fire, and can read High Valyrian, can read it, so secrecy is beside the point. What it records is a prophecy, which says nothing about who succeeds whom, and it is a family heirloom rather than a public instrument. The next day, in the carriage on the way to the crowning, the dagger is in Alicent’s hands, and she puts it into Aegon’s. A record that settles nothing and can be carried into the other camp governs nothing.

The oaths fail more instructively, because they bound some and not all. At that council, Lord Beesbury calls the crowning treason, refuses to break his oath and is killed at the council table for it. Ser Harrold Westerling, who commands the Kingsguard, hands back his white cloak and walks out. Lady Fell keeps her oath, and so does one other house, and both are dragged away by guards. Lord Caswell feigns compliance, tries to slip out for Dragonstone, and is hanged as a warning. The oaths bound individuals and did not bind the realm. Every cost of the oath fell on the people who kept it, and in that room, breaking it cost the others nothing at all. An instrument that punishes compliance and does not touch defection is not enforcing anything; it is selecting for the people prepared to ignore it.

He could have called a Great Council, put his daughter’s succession to the body that had settled the last disputed one, and pressed for the answer he wanted while he was alive to press. He did not, and the reason is plain. The precedent that Council had set was the one that buried her. Male preference was not an assumption floating in the air; it was the operative principle, and the decisive precedent in living memory. Put the question to that body after Aegon was born and it would very probably have chosen Aegon, and Viserys had every reason to expect it.

So the choice in front of him was never between having a rule and not having one. It was between his daughter and the existing rule. He had at least four ways to pay.

The first was to call that Council, take the answer it gave, and settle something on Rhaenyra in return. That would have cost her claim.

The second was to change the rule itself, so that a daughter could come before a son. There was no statute to amend; the rule was custom, made by what kings had done and lords had accepted. Changing it meant declaring a new rule of succession and then making the great houses accept it, one house at a time. The tools were the ordinary ones of a reign: marriages, offices, lands, hostages, and the breaking of the houses that refused. That was the rest of his reign spent fighting his own lords.

The third was to marry Rhaenyra to Aegon while both were children in his own household and his to give in marriage. The two claims would have become one line, and the lords would have had much less to choose between. Most of its real cost would have fallen on the two children, which is why it looked cheap, and why it belongs in the same ledger as the first scene. It was put to him, by Otto, and he refused it: the boy was an infant, the girl was of age, and she would have said no. What it would have cost him was a quarrel inside his own house, and that was the one currency he never spent. Cheap is not free, and the path he took instead was free.

The fourth was to abdicate, and put his daughter on the throne while he was alive to hold it for her. That differs from the oaths in one respect that matters. The oaths were a promise about a vacancy, to be kept or broken on the day the throne fell empty, which was also the day the man who had taken them was gone. Abdication leaves no vacancy. The lords would not have been promising anything about a succession to come. They would have been kneeling to a queen who already held the capital, the council and the treasury, with her father alive beside her. Otto could have been dismissed and the council remade under her authority rather than in place of it. Breaking an oath to an heir at a vacancy is a choice between claimants. Breaking faith with a reigning queen is rebellion. The plan that crowned Aegon needed an empty throne and a secret council to fill it. Abdication would have left it neither. It would also have done what none of the other three could. The rule he faced was older than Harrenhal, but Harrenhal was where it had last been applied, in front of every lord in the realm, and that application was what gave it its force. No Targaryen king had ever abdicated, which is part of what made it the dearest path. But a reigning daughter, with those same lords bending the knee to her while her father watched, would have been a precedent of the same weight. It could have bound them the way the Council’s had bound him. It would also have corrected Harrenhal, where the show is clear the wrong answer was given; Alicent says as much to Rhaenys’s face, that she should have been queen. It would have cost him the crown, and cost her nothing she had not already been promised.

Each of the four was expensive, in a different currency. The marriage was the cheapest to him because its cost fell on the children. Abdication was the dearest to him, and the only one whose whole cost he would have paid himself. He took none of them. He took a fifth path: he made her a personal exception and let his own presence stand in for the ratification he never sought.

That path was free, and that was its entire attraction. Free means something precise here. He paid plenty in the ordinary currency of a divided court, for 20 years. He paid nothing in the only currency that would have made the arrangement outlive him. He left the lords long an option: honour his designation after his death if it suited them, or revert to the existing rule if it did not. No premium changed hands. Rhaenyra was left short that option without ever agreeing to write it. The cost of closing it would have fallen on him; the cost of its exercise fell on her, on a date he would not be present for.

He had already shown, in the first hour, that he would let someone else pay for this problem. What he would not pay was his own peace.

The war does not begin the way I first thought it did. Viserys dies confused, speaking about the prophecy to Alicent while believing he is speaking to his daughter, and she hears her son’s name and takes it as a deathbed reversal. But when she carries it to the small council she finds they were not waiting for it. Her father Otto and the men around him had, in the council’s own words, long-laid plans for Aegon’s accession, and her account changes nothing in that room. The one man who says he does not believe it is Beesbury. The rest do not need to believe it. Aegon himself, told in the carriage on the way to his crowning, thinks his mother is toying with him.

Viserys’s wishes were not obscure. They were declared before the realm and witnessed for some 20 years, and every man on that council knew what they were. What Alicent’s party, the Greens, had was a rival principle, unrepealed and never overruled, and enough of the state in their hands to enforce it. The authority behind his answer was the thing in dispute. His death supplied the timing; the council was ready. What the misheard words supplied was cover, and Alicent’s cooperation, which was useful at the moment of coronation even though the plan had never depended on her.

That is why the correction settles nothing when it comes. Rhaenyra reaches Alicent, hears the dying words repeated, recognises the phrase from her father’s story and tells her there has been a mistake. Alicent does not accept it then; she says only that it is too late. By the end of the second season she does accept it, comes to Dragonstone herself and offers to see the gates of the capital opened. The city changes hands on the strength of that offer. What acceptance could no longer do was uncrown a king, restore the man killed at the council table and the others punished for keeping their oaths, or stop the war. She had mistaken the meaning; she had not originated the coup and was not necessary to it.

What Rhaenyra inherits is the position her father wrote and never funded.

She governs as though she has inherited a settlement, and that distinction is most of her problem. It is not for want of effort. She spent years on Dragonstone building a household and gathering allies, and she was not idle in the months before her father died. What she could not do, because it was never hers to do, was price the contest. The instruments that set the terms of the succession belonged to the reign before hers. By the time the difference between a coronation and a contest was clear to her, they were years out of reach. Her father named her, the lords swore, and for the better part of two decades his word closed the question in any room he was standing in. She arrives at her own accession expecting the realm to honour a promise made to a man who is no longer there to hold anyone to it. She is unprepared for the other side holding a rule, and the capital, and the treasury, and the council chamber, and the Faith. She has a claim, dragons, the Velaryon fleet, a handful of allied houses and an island. She lacks any instrument that could set the terms, and she was priced out of those before the bidding opened.

So she reaches for whatever is left that requires nobody else’s agreement. Each of the three can be started without asking anyone, and each sends its bill later.

The first is expedience. People held only by their usefulness are dropped when they stop being useful. Lord Corlys Velaryon, whose fleet is most of her navy, is left in Hightower hands while the queen declines to spend anything to recover him. His house comes to demand a rescue and is turned away. Baela, his granddaughter and Daemon’s daughter, accuses her to her face of using the people she needs and ignoring the rest.

The second is heaven. Aegon is revealed to be alive and the smallfolk take his survival as a sign of divine favour. The High Septon tells her plainly that this is the clearest indication of whom the gods have chosen. He refuses to anoint her. She declares him a traitor and has him killed. Then she walks out to tell the city that she is the one the Conqueror’s dream promised, the Targaryen who must hold the throne when the long winter comes. That is a claim from the prophecy, not from the succession, and the prophecy, as the dagger showed, names no heir. She is making it name her. When consent has become unavailable, a claim of divine appointment is one of the few claims left that requires none. She had other languages. She was her father’s named heir with the realm’s oaths behind her, which is the language of law. She held the city, and could have governed it and let order speak for her. She could have bargained with the Faith instead of killing its High Septon. Each was slower, and each needed somebody’s agreement. Heaven is the one she reaches for, and once she has, both sides are bidding for it.

The third is force, and the season ends on what it costs.

The order goes out in the first hour of the same panic. She tells Daemon, her husband, to take Tumbleton and root out every Hightower and kill them to the last man. She adds that the realm is watching and he should not make a monster of her. He goes on his dragon, with the army of the Riverlands and the Northern host called the Winter Wolves behind him.

Outside the walls, going behind his own commander’s back, Alicent’s brother Gwayne Hightower brings Corlys out as a bargaining chip and offers terms. Divide the realm, let Alicent’s youngest son Daeron rule Oldtown as a king in his own right, and Rhaenyra keeps the rest. Corlys, who did not know what Gwayne was going to say, improvises a better version on the spot. Keep the realm whole. Let Daeron hold the Reach under Rhaenyra’s authority. Seal it with a marriage to Baela’s sister Rhaena.

Daemon refuses where he stands, takes Corlys as his own prisoner and begins the assault.

Part of that is Daemon’s own temper, which the show has spent three seasons establishing; he kept his own judgement and he used it. But consider what he had been given. A direct order to kill every Hightower, and no instruction to do anything else. And the men across from him could not deliver what they were offering. Gwayne was acting without his commander, Corlys was a hostage improvising, and neither could bind Aegon or his brother Aemond, both alive and neither party to any of it. So the best outline of a peace the war produced was made by men who could not bind their own side. It was handed to a commander with no duty to pass it on, who was also the only road by which it could have reached the one person able to decide.

She had given Daemon force and reserved nothing to herself, not even the terms of a peace. He could have sent word back, and the decision not to is his. But nothing in her order, and nothing the series shows, required an offer of that size to reach the queen before her commander acted on it. Handing over an army is done in a sentence. Building the channel that would have brought the terms home takes forethought and costs the giver something, and she was her father’s daughter in exactly that.

From here the series stops and the written record takes over. What follows is the ending Fire & Blood gives, and then the dynasty’s later written history, not the ending the series has shown, and there is a season still to come.

In the record, Rhaenyra loses the capital, and in the confusion after she flees several pretenders hold parts of the city. She dies at Dragonstone. Aegon, Alicent’s son, the Aegon II of the record, is restored, outlives her by less than a year and is then poisoned. The record never settles whose hand it was; the men it tries for it are his own councillors and guards. (Viewers of Game of Thrones, set nearly two centuries on, will notice the shape: Joffrey too was a king his mother’s house had installed, unfit for the seat, and poisoned in it.) The boy crowned in his place is Rhaenyra’s son by Daemon, Aegon III, governed by regents, and married to Aegon II’s surviving daughter, Jaehaera, to fuse the two branches into one.

That last instrument is a marriage between the two lines, and this is the third time it appears in the story. Otto put it to Viserys when Rhaenyra and Aegon were both children in his household and every lord in the realm still answered to him. He refused it; it would have cost him a quarrel. A hostage and an envoy without authority improvised it in a field outside Tumbleton, and it was refused again. The record reaches for it only in 131, after two years of war and much of a generation, when death and exhaustion have done most of the work a settlement might once have done. By then it seals a war that has largely burned itself out.

Rhaenyra loses the throne she fought to occupy and then loses her place in the official sequence, which declines to count her as a reigning queen and runs the list of kings straight past her. What she does not lose is the line. In the later written history, that marriage produces no child; Jaehaera dies young, and Alicent’s blood, which had reached the throne in her, goes no further as the record counts it. Aegon III has sons by a second wife, two of them reign, and after them his brother, Rhaenyra’s other son by Daemon, takes the throne. Every Targaryen king who follows descends from her. No woman ever inherits the throne, and the Dance is cited afterwards to defeat every daughter who tries. It is nearly two centuries, and a different story, before a woman holds the Iron Throne at all, and only one ever sits it. She crowns herself when her last son is dead. The next takes the city with a dragon, touches the throne, and is killed before she can sit it. She is of Rhaenyra’s own line; by her time, every known Targaryen is. Nearly two centuries on, the blood Viserys preserved reaches for the same throne through another daughter. She ends as her ancestor did: the capital taken, the reign never counted, and the end betrayed and delivered by her own side. The man who delivers it gives up the rest of his life for doing so. Neither inherits. The rule that buried Rhaenyra is never repealed; it is only, in the end, taken by force. Both women are dead within weeks of each other, and the show’s last act is a Great Council. The lords meet in the ruins of the capital, choose a king, and change the rule while they are at it: no throne will pass by blood again, and the lords will choose each time. The story ends on the instrument it began with, used at last to change the rule rather than to apply it. That is the second of Viserys’s four ways to pay, change the rule, done by somebody else, nearly two centuries late. Whether those lords would ever choose a woman, the story does not say. In the same hour one of them, a woman, refuses to place her kingdom under the chosen king at all. She takes the North out of the realm, its lords crown her, and the last coronation the story shows is hers.

Viserys’s blood would have held the throne whichever side won; both claimants were his children. What survived was the line he had chosen, Rhaenyra’s, in her sons. His exception never became a rule. He got the continuity he wanted through the daughter he had named, and lost the principle it was supposed to carry, and the principle was the part that would have made her own reign possible.

Organisations run this account all the time, and the entries look the same.

There is always a version of the settlement that is free today. Name the successor without changing the mandate that governs how successors are chosen. Announce the strategy without amending the incentives that reward the old one. Promise the partner something the board has never ratified. Swear the oath once and never renew it, an instrument with no maintenance budget. Each of those is a real decision, and usually an honest one. Each prices at zero on the day it is made, because the person making it intends to be there to enforce it.

The premium is the part nobody wants to pay, and it is always the same premium: a fight with the people whose agreement would make the thing binding. Put it to the board that might say no, or renegotiate the terms with the partner who benefits from the old ones. Reopen the arrangement while you still have the standing to force it, rather than leaving it to be reopened by people who will have none. Or hand over the seat while you are still in the building, so that the handover is enforced by you and not by your absence.

Precedent is the part of that fight leaders fear most, and it is the least deserving of the fear. A precedent is only what somebody once had the standing to do and did. That is why it binds, and it is also why anyone with the same standing can replace it, at the cost of being the one who paid. Most leaders would rather inherit a precedent they dislike than set a new one. The old one costs them nothing while they defer to it. The new one costs a fight now, for a benefit that arrives after they have gone.

Laziness has little to do with why the free version is so attractive. The cost is real, the benefit of paying it is invisible, and it accrues entirely to somebody who is not in the room and cannot thank you.

Viserys decided in the first hour of the story what his wife would pay, and in the 20 years that followed he would not pay for the succession himself. The Council would have cost him years he did not want to spend. The marriage would have cost him a quarrel inside his own house, and its real price would have fallen on his children, which is what made it cheap. Abdication would have cost him the crown, and nobody else anything. He paid none of them. The fight he declined in politics came back as war. The payout fell first on his daughter and the people who had kept faith with her. Then it fell on a realm that, by the record’s count, spent two years and much of a generation buying a settlement he still had the standing to compel.

So the questions to put to oneself are these. What have I decided that nobody with the power to overturn it has ever agreed to? What holds only because I am there to hold it? Where do the people who keep faith with me carry a cost for it, while the people who break faith pay nothing? What have I delegated without saying what must come back before it is decided? What have I wanted for so long that I have stopped counting what taking it costs me and the people bound to me? And what is the fight I keep deferring, what would it cost me to have it now, and who pays for it if I never do?

Have a reflective weekend.

AI-illustration: A results portal at zero of 226, and police tape across a courthouse door.

Part 1, Before the Vote, priced what the 13 August ballot tested. Part 2 reads the answer. It is the final entry in this lane; if the Constitutional Court acts on the documents before it, a dated addendum attaches here, and the economy moves to the September fiscal note.

Part 1, “Before the Vote” (Canary Compass, August 2026), put the election inside one system with two channels. Policy passes down through banks, prices and credit to households, or bypasses them by direct delivery that lasts as long as it is financed. Information passes up from citizens through speech, the press and assembly, including what the state does not want to hear, and without it the state receives only confirmation. Part 1’s scope condition was that stabilisation could be delivered without the second channel and that diffusion and integrity could not. Its rule was that the incumbent is assessed on state power exercised and challengers on record and conduct. Part 2 was to print on the results and score that test. The count that would let it do so did not come, and the fortnight that followed is the subject instead.

Zambia declared a winner and published no written constituency count. The courts were sealed by police from the morning of the 24th to the 26th, across the last day a citizen could contest the result. They reopened the day after the Judiciary declared no petition had been filed. The downward channel, policy reaching households through banks and prices, is intact and legislated to widen. The upward channel, narrowed by statute over the past year as Part 1 set out, still carried more than usual this fortnight. Observers, the press and citizens reported everything set out below, and the state heard it. What closed were the two points at which what citizens know can bind the state: the written count that would let the declaration be checked, and the courtroom that could act on the check.

1. What Is Settled and What Is Not

1.1 The Count

In the fortnight after the vote, opposition members were arrested, a former Cabinet minister, Mutotwe Kafwaya, was shot dead during a security operation, the count was suspended, and the courts were sealed. None of those settles the result.

On 14 August, the day after the vote, the Electoral Commission of Zambia suspended the counting and announcement of results throughout the country for about six hours under section 67(3) of the Electoral Process Act. It cited violence against poll staff and the theft of marked ballots. On the 18th it declared President Hakainde Hichilema the winner. The Commission read constituency figures aloud on air and declared national totals from them, and the announced figures reconcile to themselves. But the Commission has published no written record of them. The portal it built for every constituency, and launched the day before the vote, still shows zero of 226. There is no published dataset of votes cast, presidential and parliamentary, valid and rejected, across the 226 seats. It is the plain constituency results the Commission's own officers read out, put in writing on the Commission's own website. Part 1 asked that the chain from station to declaration stay public. The ask since the Friday Reflection is narrower, and the Commission has not met the narrower one. Without it there is no complete official record of the votes cast to examine.

1.2 The Remedy

Article 103 allows seven days from the declaration. The Judiciary’s own letter to Secretary to the Cabinet Patrick Kangwa placed the declaration between 02:00 and 06:00 on 18 August, so the window ran to the same hours on the 25th. Under Order XVII rule 8 of the Constitutional Court Rules, Statutory Instrument 33 of 2026, the registry sits from 08:30 to 18:00 and does not sit before dawn. The last hours in which a petition could be lodged therefore fell within the working day of the 24th. Under Order III an electronic filing is made through the Master’s filing system, with a user identity, an acknowledgment and a stamp, and an email to the Registry is not one. The system those rules describe was still in development at the last public statement on it, a contractor’s notice published in January 2025 and last updated that August. No court has announced it in service since.

On the morning of the 24th the superior courts in Lusaka were sealed by police and heavily armed military personnel, as the Law Association of Zambia recorded. The Judiciary’s Chief Administrator, Nalishebo Imataa, told staff by memorandum not to report, for security reasons, with guidance to follow once the situation was contained. The Law Association saw the memorandum unsigned and recorded that no further explanation was given. Police tape marked some premises as crime scenes. The police described the closures as intelligence-led and preventive security assessments at several government premises, without detail. State House said the President had neither ordered the closure nor any power over what the courts receive. That day the lawyer Linda Kasonde, a former Law Association president, emailed a petition to Chief Justice Mumba Malila after, she says, a message to the registry’s address bounced. On the last filing day the physical route was sealed, and no public record shows the electronic route existed.

On the 25th the Judiciary’s letter to Kangwa reported no petition filed in the Registry as at 18:00 on the 24th, counting the window exactly as above. Kangwa announced that the swearing-in proceeds on 1 September. The same day the Judiciary confirmed that Malila had referred the documents to the Constitutional Court on the question of whether they constitute a filing. He presides over the Supreme Court, not the Constitutional Court, which alone hears a presidential petition. That is the head of one apex court passing papers to the separate court that would rule, not the Constitutional Court receiving a petition. On the 26th the Judiciary, through its administration, closed the question. The Registry had received no Presidential Petition as at the close of business on the 25th, and so, in its own words, there is none before that Court. This is the administration reporting an absence; the Constitutional Court itself has ruled on nothing. That evening the Chief Administrator told staff that, “following the de-escalation of the security threat”, they could return to work on the 27th. The courts had been sealed for three days. The security threat named in the memo appeared on the last day a petition could be lodged and was gone once it could not.

1.3 The Attempts

The alliance says it holds signed forms and tendered copies to the observer missions, and its candidate filed nothing. Kangwa’s statement on the Kabulonga operation said the opposition candidate Brian Mundubile was found at the scene; Mundubile says he was detained and released the same day. From 17 August he and his running mate Makebi Zulu were sheltering at United Nations premises in Lusaka. On 20 August the police issued a call-out for both over treason and insurrection allegations, which they deny. His camp says lawyers and officials assembling the case were arrested; that is not verified. The governing party said he was free to use the legal routes available, and the government says the operation concerned a militia and not the petition. Kasonde had told stakeholders the day before that efforts to get a citizens’ petition off the ground had not succeeded, then emailed one anyway. It cites the observer network’s estimate against the declared share, the tally-centre discrepancies and the Kabulonga killing. Whether those grounds would have held is not before any court, because no court has taken the petition up. No timestamped record of the attempt has been produced by either side. Kasonde faced no allegation and could act from outside the building. The candidate was in shelter under a summons, with no registry open to his lawyers.

A presidential petition is built in seven days from the station record, the signed result forms held by party agents and observers. Between the vote and the declaration that record was degraded at the stage where it is consolidated. The European Union mission rated tabulation very good in 43 per cent of its observations before the six-hour suspension on the Friday and found it deteriorated significantly after it. It recorded a military presence at totalling centres that left fewer party agents and citizen observers in the room. An armed group took almost all the result forms from the Makeni totalling centre on the Saturday. The record was thinned by the deployments the mission observed and by the Makeni seizure, whose perpetrators are not identified. Whether it was thinned enough to stop a petition, no petition tested.

1.4 What Hardened and What Did Not

The reconciliation is set out in “Polished Teeth, Blank Portal” (Canary Compass, August 2026). The reading has hardened across three dates. On 12 August the note priced an offer and a record. On 21 August the reconciliation ran into a wall the Commission built, with no figures published in writing to check against. On 24 August the remedy closed, and on the 25th the state proceeded on an administrative letter. The verdict on the count itself still waits for the record. What has hardened is not a verdict on the count but the reading of what the state did around it.

What is not settled is the count itself, and it cannot be, on the present record. A declaration is a claim; the published dataset is the document that would test it; Zambia has the first without the second. Others have reached for a forensic audit of the polling-station forms. That sits a level above where this stands, because the layer beneath it is empty. No swing, no share, no provincial reading can be built from a published record that does not exist. The inauguration proceeds on 1 September with none before the court.

2. The Dashboard Has a Second Term in It

At end-May, on the Ministry's cash figures against a 2026 GDP of K821bn, the overall fiscal balance stood at minus 2.15 per cent and the primary balance at plus 0.2. That GDP is the 2025 preliminary outturn grown at the International Monetary Fund's May projections for real growth and inflation, and it sits a ninth below the K924bn the budget assumes. Part 1 priced the direction. The primary surplus was 3.1 per cent in 2025 and the budget's K39.2bn target for 2026 is 4.8 on the same base. The Fund's May projection was 1.1, and at five months it is 0.2. Debt service did not do that; interest ran below a straight twelfth of its annual line, the measure used here for every line beneath the totals. Revenue did, 14 per cent under its monthly targets, with value added tax (VAT) collecting half its twelfth. Grants ran at a quarter of theirs on the monthly returns, a third once an unattributed K458m appears in the cumulative line. One spending line did, transfers and subsidies a quarter over on the agricultural lines Part 1 called the bypass bill, measured before the April reclassification of K2.8bn of input-support spending to assets. Total spending was 5 per cent under the Ministry's own monthly profile at five months, and the shape matters more than the average. January ran a fifth over profile on interest and capital spending, and March 41 per cent over on wages and the transfers month. Then April, the largest month in the profile, came in a third under and May a fifth under. The pullback came in the two months before the campaign's climax, and revenue fell faster than spending was held. The deficit at K17.7bn had used nine tenths of the year's original K19.4bn room by May. Against the budget's own GDP that room is the 2.1 per cent of the budget speech; against the K821bn it is 2.4. The outturn, at 2.15, had already passed the budget speech's ratio for the year. The supplementary estimates approved on 11 May added K26.3bn of spending, K16.3bn net of rationalisation, and the revised deficit is still unpublished. External programme money is barely moving, and net external financing at end-May was negative once amortisation is counted. In an election year the state is a net external payer, and the deficit is carried at home.

Copper is the offset, and the royalty line qualifies it. Over the five months the royalty line covers, the London price set its record in January and posted in April its second-highest monthly average since 2014. Mineral royalty over the same months stood at 97 per cent of its twelfths, on pace and no more. A record dollar price collecting a budget-pace kwacha royalty is the exchange rate at work, the trade-off set out in “Copper Output and the 2026 Royalty Arithmetic” (Canary Compass, January 2026). August's prices, London within sight of the January record and New York at records, are the second half's revenue question, not the first five months' answer. On the Bank of Zambia's own real effective exchange rate index, a lower reading is a stronger kwacha. In May the kwacha was at its strongest level in the series since 2003, a third below its long-run average. The kwacha spent the election month between 18.70 and 19.05 on the interbank daily average, inside the 18 to 20 band that suited stabilisation, when the price level was the priority. With the vote behind it, the priority changes. The Bank has room to buy reserves into the offshore inflow and let the kwacha drift, 20 to 22 first and 22 to 24 over the medium term. That lifts kwacha revenue per tonne and gives the returning offshore bid a cheaper entry into bonds whose seven-year yield fell 101 basis points between the June and August auctions. The drift is not free on the budget's own lines. The budget carries K21.7bn of external interest and principal, about K17bn of it still to pay after May on the amortisation paid to date and interest taken pro rata, so every 1 per cent the kwacha loses adds about K170m to it. The same 1 per cent adds about K110m to the K10.8bn of royalty still to collect and about K80m to the mining company tax line, budgeted at K13.2bn. Together they only bring the two sides level. What makes depreciation pay is import VAT, K42.5bn in the budget, because the tax is charged on the kwacha value of imports. A weaker kwacha raises that value, the same rate collects more, and the consumer pays it in the price of every imported good. The fiscal gain from a weaker kwacha and its inflation arrive through the same import prices. The cost sits in the price level. The 6.5 per cent print is held down by fuel relief that expires on 30 September, the day the Monetary Policy Committee announces. The relief ends ahead of the months from November to April when inflation seasonally rises. The World Meteorological Organization (WMO) expects a strong El Niño through October. The US National Oceanic and Atmospheric Administration (NOAA) gives a greater than 90 per cent chance of a very strong event this winter. NOAA also gives a 69 per cent chance that the October to December season exceeds every El Niño since 1950 on its index. The September fiscal note takes that arithmetic up in full.

The market read the week and moved on. Table 1 sets three of the year’s six bond auctions, the first, the last before the vote and the first after it, against the kwacha on the day of each. The February auction covered 5.08 times and April’s 0.43, and that arc is among the September note’s subjects. Since the January auction every tenor has fallen, the seven-year by 206 basis points, and the cover has gone from 2.40 times to 0.77 in June and 1.37 in August. The kwacha opened the year at 22.07 to the dollar and stood at 18.90 on the August auction day. Bloomberg read the August auction as a bet on continuity, and Citi had published a buy recommendation two days before.

There is a second reading the auction covers do not rule out. By Circular 10 of 2026, dated 31 July and effective 3 August, the Bank cut the statutory reserve ratio on kwacha deposits from 26 to 21 per cent. Foreign currency stayed at 26. That was ten days before the poll and between meetings of the Committee. Governor Denny Kalyalya had announced it at the Agricultural Show on 31 July, and compliance moved from a daily test to a weekly average. Statutory reserves fell by about K6.5bn in the first week of August. Central bank term deposits rose first, then the bids at the bill sale of 6 August and the bond of the 21st. The August tender sheet splits by tenor, not by holder. The two-, three- and seven-year lines covered 1.6 to 1.8 times; the ten- and fifteen-year covered 1.2, though their bids nearly doubled from June. A plausible reading is that the freed reserves sat at the front, where banks hold, and that the long end carried the returning offshore bid Bloomberg described. Whether continuity or the release was the larger draw is not settled by the covers. The October monetary statistics show the allocation: whether private kwacha credit grew, or bank holdings of government paper did.

The design of the release is on the record. “Zambia Monetary Policy, Part 3” (Canary Compass, December 2025) proposed the instrument, conditioned. Kwacha reserves down, foreign currency reserves released in kwacha, and the relief tied to private lending, so the freed liquidity could not simply return to government paper. The kwacha leg arrived alone and unconditioned. The default use of unconditioned reserves in this system is government paper, and within three weeks the bids for it had risen. The condition that tied the relief to private lending did not arrive with it, and whether the freed liquidity ends in private credit or in government paper is the allocation question above.

That matters because the economic programme legislated for the term the declaration awards is transmission. The April legislation, and the rules issued under it, carry it. The Banking and Financial Services Act with its netting provisions, the Agricultural Credits and Warehouse Receipts Act and its companion, the currency directives, the deposit insurance framework, the concentration tools in the capital rules. They exist to move bank balance sheets out of sovereign paper and into households and firms. A state that means that cannot also fund itself by deepening the banks’ holdings of its own debt; for a given deposit base the two are the same balance sheet. Two monetary measures followed that legislation, a 25 basis point cut in the policy rate on 13 May and the reserve cut of 31 July. Only the second put liquidity on the banks’ balance sheets. The demand it visibly fed, on the auction record, was for government paper; whether it stays there is the allocation question above. If the banks are to be pushed the other way, the deficit has to shrink or be funded from outside them. Mining growth under the 3 million tonne programme does the first, in time; external markets and concessional bilateral money are the second. In July Situmbeko Musokotwane, the former Minister of Finance, told Reuters the government wanted a new Fund arrangement by year end. On radio on 25 August Musokotwane said the need for a Fund programme will disappear within five years, and gave mining growth as the reason. The two are one position: the mines close the gap eventually, and something funds it until they do. Citi’s own note says there is no external return before a programme, so for those years the outward route runs through the arrangement the July request asked for. Until then the only lender available is the one that legislation is written to release.

A reader of Part 1 asked whether the narrowing of the upward channel extends to sovereign credit. The 2033 bond is the gauge. Its yield rose from 6.52 per cent on 20 July to 6.69 on 30 July and held near 6.65 to the eve of the vote. The plausible reading is that the widening tracked the late lift in Mundubile’s crowds and the probability of a run-off. It was 17 basis points, not a repricing, because the incumbent stayed the favourite throughout. For the market the worst case was an opposition win, the best an outright Hichilema win, and a run-off sat between. Our pre-vote forecast ordered it differently: a run-off at 50 per cent, Hichilema outright at 35 and Mundubile at 15. Had the market carried that ordering, a run-off as the modal case, July would have been a repricing. Seventeen basis points says it never did. The yield fell from the declaration, the claim rather than the count, 6.64 on the 18th, 6.55 on the 24th as the courts were sealed, 6.47 on the 26th. On that reading the market priced a contested result into the vote and unwound it on the declaration. What the series shows without a reading is that the yield fell through the days the courts were sealed. The sealing was not priced as a risk. The external market prices violence and fundamentals, and the quality of the political space enters only when it produces one of them. Across 27 emerging markets between 1996 and 2009, Eichler found that governance quality and political stability move sovereign spreads while the degree of democracy and elections do not. Vaaler, Schrage and Block found spreads rise as a right-leaning incumbent looks more likely to lose to a left-leaning challenger. That is the effect the July widening resembles, with a market-friendly incumbent and a challenger from the Tonse alliance, which the market read as the reverse. The last time Zambia was lent to on the dashboard alone, the arrears were insufficiently disclosed until 2020, and the market priced the failure only once the numbers surfaced. The infrastructure that would surface the next one is the upward channel, and the state has spent the year narrowing it. On the same reading, a market lending into an unpublished count at its tightest yield since June is pricing a promise, and the instruments that would report the promise breaking are the ones just closed. The measure to watch is the spread at the first external issuance, against the 9.14 per cent replacement threshold set in “The Most Expensive Recovery” (Canary Compass, June 2026).

3. What the Upward Channel Is For

Transmission is the hardest thing a state can attempt in an economy shaped like Zambia’s, because success is measured at the household and the state cannot see the household directly. Its administrative returns and its bank supervisors report what its own agents recorded. The upward channel is where it learns what they did not. Whether a cheaper loan reached a trader, or a result posted at a station survived to the tally centre. Full-file credit reporting and savings-group ledgers feeding the bureau are the plumbed part of the same disclosure.

Part 1 promised to read the result through this framework. Any analysis of a count starts with validation: the full set of observations, a consistency check on each, the failures flagged and set aside, and a test that what remains still represents the whole. That is the sequence the election forensics literature sets out, and its authors are explicit that verdicts wait for completed counts. Part 1's indicators, including the Constituency Development Fund (CDF) test, which could otherwise be run on the parliamentary vote, sit downstream of that step. Here the full set does not exist. The reconciliation compiled both ballots for 101 seats and could verify 40 to the ballot. Of those 40, 21 do not reconcile, presidential ballots exceeding parliamentary by 10 to 48 per cent, every one in a seat the President carried, the largest gaps in three provinces. The ordinary explanations, voters declining the second ballot, different rejection rates on the two, a shortage of one ballot, each leave a line in the station accounts. Mapatizya's parliamentary declaration reconciles to the ballot exactly and is still 10,154 short of its presidential count. Which explanation holds can be checked only against accounts that are not published. Data missing for a reason tied to the outcome are, in Rubin's classification, not missing at random, and a sample selected that way carries the bias Heckman identified. The selection argument rests on the pattern inside the verified 40; the other 186 are absent for one reason, that nothing was published, and that reason cannot be tested until it is. Inference from what remains is not valid, and excluding the failures would make it worse. The reconciliation, the observer network's parallel count and the academic turnout screen have each run a consistency check on what is available. That is validation, not the analysis it precedes. The observer missions, the opposition and the government's defenders abroad do not dispute that the figures are contested; the Commission and State House do; and no one can test them against a published record. The second term begins with its own count unexaminable.

There is a benevolent version of governing from authority, and it has a test its admirers would accept. The government is broadly accepted, it does not act afraid of its opponents, and when challenged in court it goes to court and wins. A state confident of its count publishes it and meets the petition in the courtroom. This one did neither. Zambia changed government at the ballot most recently by electing this President over an incumbent whose handling of the courts and the register was part of the case against him. A rule set aside for an administration one trusts is set aside for the one that follows, chosen by the same electorate. The economic case for a second term stands on its own numbers. The case for making it from behind a cordon does not, and both are read in 2031.

4. April Became August

Zambia’s fortnight is not confined to Zambia. “The Forced Choice” (Canary Compass, February 2026) set out the mechanism, and the fortnight is that mechanism landing. The global order is in rupture rather than transition, and in a rupture the great powers set terms and the states in between take them. Washington now selects partners on supply-chain alignment rather than on governance, through the mineral-security architecture built across late 2025 and early 2026. Beijing tolerates strategic ambiguity because the standing arrangement favours its positions. A government that pivots is offered statements rather than protection. India accepted the terms on 2 February, committing to move off Russian crude and toward American purchases to bring a tariff from 50 per cent to 18. When the Iran war closed Hormuz in March, Washington itself waived the Russian-crude condition, and the terms moved with the great power’s needs rather than India’s. Hichilema told the diplomatic corps on 29 January that the country was too small to be dragged into the geopolitics. The paper named the elections of 2026 and 2027, in Zambia, Kenya, Nigeria and South Africa, as the windows in which governments able to make structural commitments might be replaced by governments that cannot. Under that pricing, the external cost of an unpublished count and a sealed courthouse is close to nothing, and August is the forecast landing.

The defence offered this month was the comparison. Criticism from abroad was met with the critic’s own country’s record, and criticism at home with the record of the last government. Part of that defence is true and should be conceded. Some of those commenting on Zambia from abroad are quiet at home, some because speaking at home costs them something and some because it is easier to be brave about another country. A critic who will not read his own country’s record has no credibility reading Zambia’s. But measuring the country against its worst neighbour sets no standard it has to meet. The standard that binds is the one Zambia chose.

It ratified the African Charter on Democracy, Elections and Governance in 2011, a decade before Kenya in 2021 and Zimbabwe in 2022; 39 states have. Those states bound themselves to each other, not to a foreign audience, and the obligation runs both ways and runs first at home. The Charter is short and specific. Article 17 binds each state to independent and impartial electoral bodies and to national mechanisms that redress election disputes in a timely manner. It also requires a code under which results are accepted or challenged through legal channels only. Article 2 commits it to protect the independence of the judiciary, Article 3 to the separation of powers, and Article 14 to constitutional civilian control over the armed and security forces. Article 23 lists what counts as an unconstitutional change of government, and the list includes an incumbent’s refusal to relinquish power after a free election and any revision of legal instruments that infringes democratic change. Article 46 leaves the measures to the Assembly and the Peace and Security Council. The Union has not enforced it. Unenforced treaties fall to the citizens of the states that signed them, who may read them back to their governments.

In April, South Africans met African criticism of the vigilante campaign against migrants by telling other Africans to mind their own countries. In August, Zambia’s defenders gave the same answer. “Crime vs Crime” (Canary Compass, June 2026) argued that a continent which treats a neighbour’s governance as foreign business, rather than shared business, withdraws a goodwill no trade agreement can legislate back. A Ghanaian asking how Zambia reconciled its ballots is not proposing to govern Zambia. He is reading a charter Zambia signed. The Mo Ibrahim Foundation read it the same way on 26 August: the closure of the top courts “just before the deadline for challenging the election result, only adds to the mess”.

The external pressure that once met such moments has thinned from both directions, which “The Forced Choice” priced and this fortnight confirms. The first independence generation lived through a harder version, over 80 coups and attempted coups between 1960 and 1982, before the multiparty ballot returned in the 1990s. This generation is more connected and harder to govern by decree. The next tests are on the calendar, Kenya’s in August 2027 among them, and each will be defended by pointing across the nearest border. Each will be judged by two questions: whether the state publishes the count it declares, and whether a citizen who doubts it can reach a courtroom to say so. On both questions this fortnight, the answer in Zambia was no. South Africa in April became Zambia in August.

5. What Comes Next

A fiscal note in September, ahead of the Committee, reads how the year’s deficit room was used and what the supplementary estimates do to the ratios. The Committee announces on 30 September. The tell is whether the rate is set on an inflation print held down by the fuel-tax suspension, the excise and VAT relief extended to 30 September. The October monetary statistics show whether the August release reached borrowers or bonds. The first sittings of the new Parliament and the Bank’s directives under the April Acts show whether the transmission legislation is commenced or left on the statute book. And the Constitutional Court, the only court that can hear a presidential petition, has not acted on the documents referred to it; the Judiciary reports nothing before it. If that changes, a dated addendum follows.

Both channels were named before the vote. After the count, one channel was legislated wider and the other was left open to speak and closed where it binds, by the same state in the same season. The economic programme the state has legislated will be read through the channel it has shown it need not answer. It read the count aloud and has not published it; the one court that could test it has not been reached.

Sources

African Union, African Charter on Democracy, Elections and Governance, treaty text and status list, au.int/en/treaties/african-charter-democracy-elections-and-governance (Addis Ababa, adopted 2007).

Amnesty International, statement on the killing of Mutotwe Kafwaya and the arrests of 14 August 2026 (London, August 2026).

Bank of Zambia, Circular No. 10 of 2026 on statutory reserve ratios; Commercial Banks’ Liquidity fortnightly; auction results 01/2026/BA, 05/2026/BA and 06/2026/BA; interbank daily average mid rates to 25 August 2026; real effective exchange rate index to May 2026; Monetary Policy Committee calendar (Lusaka).

Bloomberg, reporting of 22 July and 24 August 2026, and the ZAMBIN 5.75 per cent 2033 historical price table, mid yield to convention, to 26 August 2026.

Canary Compass, The Last Open Channel, Part 1: Before the Vote (12 August 2026); Friday Reflections: Polished Teeth, Blank Portal (21 August 2026); The Forced Choice (February 2026); The Most Expensive Recovery (June 2026); Crime vs Crime (19 June 2026); Zambia Monetary Policy, Part 3 (December 2025); Copper Output and the 2026 Royalty Arithmetic (January 2026); and author’s calculations.

Eichler, Stefan, The Political Determinants of Sovereign Bond Yield Spreads, Journal of International Money and Finance 46, 2014, 82–103.

Electoral Commission of Zambia, press statement of 14 August 2026 on the suspension of counting and announcement of results; declaration of the presidential result of 18 August 2026; results portal, results.elections.org.zm, checked 26 August 2026 (Lusaka).

European Union Election Observation Mission to Zambia, Preliminary Statement of 15 August 2026 and Election Observation Update of 19 August 2026 (Lusaka).

Government of Zambia, Constitution of Zambia (Amendment) Act No. 2 of 2016, Article 103; Constitutional Court Rules, Statutory Instrument No. 33 of 2026, made 11 May and commenced 15 May 2026, Orders II, III, XVII and XVIII; Statutory Instruments Nos. 56 and 61 of 2026 (Lusaka).

Heckman, James J., Sample Selection Bias as a Specification Error, Econometrica 47(1), 1979, 153–161.

Hicken, Allen and Walter R. Mebane, Jr., A Guide to Election Forensics, USAID Research and Innovation Grants Working Paper, 2015, iie.org/publications/dfg-um-publication; Mebane, Walter R., Jr., Inappropriate Applications of Benford’s Law Regularities to Some Data from the 2020 Presidential Election in the United States, working paper, University of Michigan, 2020.

Judiciary of Zambia, Chief Administrator’s memorandum of 24 August 2026 on the closure of court premises and letter to the Secretary to the Cabinet of 25 August 2026, as reported; press statements of 25 and 26 August 2026; Chief Administrator’s memorandum of 26 August 2026 on the reopening of court premises (Lusaka).

Kasonde, Linda, communication to stakeholders of 23 August 2026 and letter to the Chief Justice of 24 August 2026, as reported.

Law Association of Zambia, statement of 24 August 2026 (Lusaka).

London Metal Exchange and COMEX, copper settlement prices, January to August 2026; World Bank, monthly copper price series to April 2026.

McGowan, Patrick J., African Military Coups d’Etat, 1956 to 2001, Journal of Modern African Studies 41(3), 2003; Powell, Jonathan M. and Clayton L. Thyne, Global Instances of Coups from 1950 to 2010, Journal of Peace Research 48(2), 2011.

Ministry of Finance and National Planning, Monthly Economic Indicators, January to May 2026; 2026 Budget Speech and resource envelope; Supplementary Estimates No. 1 of 2026 (Lusaka). GDP base for ratios: Zambia Statistics Agency 2025 preliminary nominal GDP of K725.6bn grown at the IMF May 2026 projections of 4.3 per cent real growth and 8.5 per cent inflation, K821bn; the budget’s implied GDP of K924bn is noted where used.

Mo Ibrahim Foundation, statement on Zambia of 26 August 2026 (London).

Mundubile, Brian, statements of 17 to 23 August 2026 and broadcast interviews, as reported.

Mwamba, Emmanuel, letter to the Chairperson of the African Union of 23 August 2026, as reported.

Press reporting: Africanews, Bloomberg and CNBC (US waiver on Russian crude for India, March to May 2026), Business Day, Lusaka Times, News Diggers (No. 2328, 26 August 2026, reporting the former Minister of Finance’s 5FM Radio interview of 25 August; and the Chief Administrator’s letter of 25 August, as quoted), Reuters, and the Zambian Observer.

Rubin, Donald B., Inference and Missing Data, Biometrika 63(3), 1976, 581–592.

Secretary to the Cabinet, statements of 15 and 25 August 2026 (Lusaka).

State House, statement on the closure of the courts, 25 August 2026; Zambia Police Service, statement of 24 August 2026 (Lusaka).

Synergy International Systems, Transforming Justice in Zambia with Electronic Case Management System, synisys.com/news/transforming-justice-in-zambia-with-electronic-case-management-system, published 20 January 2025 and updated 5 August 2025.

Vaaler, Paul M., Burkhard N. Schrage and Steven A. Block, Counting the Investor Vote: Political Business Cycle Effects on Sovereign Bond Spreads in Developing Countries, Journal of International Business Studies 36, 2005, 62–88.

World Meteorological Organization, Global Seasonal Climate Update and statement on El Nino (Geneva, August 2026); NOAA Climate Prediction Center, ENSO outlook (August 2026).

ZambiaLII, Acts of Parliament Nos. 6, 7, 8, 9 and 11 of 2026 (Lusaka).

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: Two ballots issued to the same hand at the same station.

A state with more holes in its story than a 1960s rock star’s jeans.

A dead former cabinet minister. A permanent secretary who dismissed reports of his shooting, advised Zambians to “be serious,” and was contradicted days later by his own police. An official presidential result that the country’s largest observer network says sits outside its estimated range, with votes added in 30 constituencies. A Constitutional Court expected to hear any petition after three judges were removed in 2024 and three new judges were appointed in May 2026. An armed insurrection involving five AK-47s, 60 rounds, and other weapons, for eleven people now held for treason-related offences. Five rounds per plotter, with five to spare. An Electoral Commission that declared the result at three in the morning on 18 August and, three days into the seven-day window the Constitution allows for challenges, has still not put the figures on the results website it launched the day before the vote.

All this in 2026.

I am not convinced any of it is real. It reads like the plot of a Bollywood action film, and if you have sat through one you know exactly what I mean.

But of course, raising concern is a foreign narrative, manufactured by coloniser minds determined to make Africa look bad.

I wrote The Silence Beneath the Canopy in May 2025. It is about what happens to a place when the questions stop. Go and read it.

Mutotwe Kafwaya is dead. He was 49, the former member of parliament for Lunte, and under President Lungu the Minister of Works and Supply and then of Transport. Whatever else is written this week, a man has died, and nothing below is meant to make that lighter than it is.

Reports said he had been shot during a security operation and taken to Maina Soko. On 16 August the government called the reports a total fabrication, and the Defence Permanent Secretary, Maambo Haamaundu, told Zambians that “we need to be serious as a people.” His wife asked the Lusaka High Court to hear urgently an application over what she called his continued detention. The Human Rights Commission said it could not establish whether he was in state custody. On 19 August the Inspector General of Police confirmed that Kafwaya was the previously unidentified man shot during the joint operation in Kabulonga on the evening of 14 August, taken to hospital, and pronounced dead. His family identified the body.

Governments hold back national security matters, sometimes rightly, and the courts will hear what happened in that house. None of that required a denial. Five days passed between the shooting and the confirmation, with his wife in the High Court asking where he was and the state telling the public to be serious.

Eleven people were arrested that night at a Kabulonga house used by the main opposition candidate. High-grade military weapons, said the Secretary to the Cabinet. The police say the individuals had been under watch since November 2023. Nearly three years of surveillance, and nobody noticed the coup was short of bullets.

When the announcements started, I posted that one of the first tests of an election is to set presidential votes beside parliamentary votes cast on the same day. I was not looking at anything in particular. What surprised me first was Lusaka and the Copperbelt, which looked more competitive than I had expected. Then someone pointed me to Solwezi Central and Mongu Central and asked whether I had seen the gaps between the two ballots. I assumed outliers and checked more.

The test: voters received both ballots together, at the same station, in the same visit. Within a constituency, the two totals must track. In 2021, on the ECZ’s own figures, they tracked nationally to within 0.31 percentage points. In 2016, to within 0.42. I compiled both races for 101 constituencies from ECZ read-outs as transcribed by several outlets, returning-officer declarations, and party and network collections, keeping every source and every conflict. Media transcriptions contain errors, and the ECZ has published nothing in writing, so only 40 of the 101 could be verified to the ballot, meaning candidates plus rejected equals the declared cast or the cast total is directly published. Everything below rests on those 40.

Nineteen reconcile within 10 per cent, seventeen of them within 5. In Southern: Monze East, Itezhi-Tezhi East, Namwala West. North-Western: Zambezi Central. Western: Mitete. Copperbelt: Kabushi, Bwana Mkubwa, Konkola, Kalulushi, Chingola Central, Roan. Lusaka: Chilanga South, Matero. Central: Kapiri Mposhi West. Luapula: Chifunabuli South, Nchelenge. Eastern: Chipata North. Northern: Mpulungu North. Muchinga: Nakonde. Seats the President carried and seats he lost. Twenty-one do not reconcile. Presidential ballots exceed parliamentary by 10 to 48 per cent, 186,247 ballots in total. North-Western: Solwezi Central, Solwezi North, Mufumbwe, Manyinga, Mwinilunga East, Mwinilunga West, Kabompo East, Kabompo West. Southern: Mapatizya, Mazabuka Central, Kalomo Central, Chikankata, Magoye East. Western: Mongu Central, Luena, Senanga South, Lukulu South. Copperbelt: Luanshya, Lufwanyama East, Lufwanyama West. Lusaka: Roma. Mongu Central: 50,618 presidential ballots, 30,531 parliamentary. Turnout 90 per cent on one ballot and 54 on the other.

All 21 were carried by the President. No verified constituency carried by the opposition shows any excess. The largest divergences cluster in North-Western, Southern and Western, which is where everyone expected the President’s heaviest support. But the cluster is not clean. Inside it, seats of the same character close and do not close. Monze East, Itezhi-Tezhi East and Namwala West in Southern, Zambezi Central in North-Western and Mitete in Western reconcile. Mapatizya, same province as Monze East, same candidate at 100 per cent of the two-candidate vote, is missing a quarter of its parliamentary ballots. Outside it, the Copperbelt holds six seats that close and three that do not, and Roma in Lusaka does not. So “strongholds behave differently” explains the provinces. It does not explain Monze East against Mapatizya, Zambezi Central against Solwezi Central, or Kabushi against Luanshya.

An election is a closed accounting loop. Every polling station records ballots received, issued, in the box, rejected, and unused and returned. A voter who takes the presidential ballot and declines the parliamentary one leaves a parliamentary ballot in the unused pile, or a blank one in the box, or a rejected one in the count. A vote for an independent is a parliamentary ballot cast and is already inside these totals. Mapatizya’s parliamentary declaration reconciles to the ballot exactly, 29,142 plus 528 rejected, and it is still 10,154 short of its presidential count. So “they chose not to vote for MPs” is a line item, and it is checkable. Abstention, differential rejection, a ballot shortage, transcription error, additions at the tally centre: each leaves a different mark in the accounts. Anyone naming the cause before the accounts are published is guessing.

CCMG got there by a better road. Its observers sat in 1,491 sampled polling stations from opening until the official station result was posted, and recorded that result. Its PRVT estimated turnout at 53.5 per cent plus or minus 0.6, the incumbent at 56.0 plus or minus 1.7, Mundubile at 42.3 plus or minus 1.7. The official figures for all three fell outside those ranges, because in 30 constituencies official turnout sat more than 10 points above the plausible upper limit, driven by increased UPND votes. In four constituencies where CCMG obtained tallying records, it compared 24 sampled stations. At ten, UPND votes were higher at the tally centre than at the polling station, by an average of 178. CCMG also says the incumbent probably crossed 50.

Nic Cheeseman and Nicole Beardsworth found 27 constituencies with presidential turnout above 80 per cent: nine in North-Western, eight in Western, six in Southern, three in rural Copperbelt, one in Central, none in the four provinces where the opposition did well. Solwezi North: a gap of more than 46 points between the two ballots.

One explanation offered is a flawed adoption process: voters so unhappy with the parliamentary field that they refused the second ballot, but not so unhappy that they stayed home. Adoption disputes usually produce independents, and angry voters vote for them. If the unused and rejected parliamentary counts for Mongu Central and Solwezi North account for the gap, adoption becomes one explanation worth testing among others. Nobody can get to that question until the counts are published. The question on the table is the one before it.

Every serious way of checking an election ends at the official figures. Forecasts, sample-based estimates and interviews with people who were in the room are inputs waiting for the thing they are measured against. A forecast that lands on the announced number has matched the announcement. Whether it has matched the election is the question.

The EU Election Observation Mission has observed every general election since 2001 at the government’s request. At ten of the 38 polling stations where its observers watched the count, the ballots in the box did not match the ballots issued. On 14 August the ECZ suspended counting and announcements for six hours. The mission’s update of 19 August says that after the suspension, observers recorded extended breaks in almost half their observations, with returning officers appearing to wait for instructions from ECZ headquarters before announcing results. A heavy military presence thinned out party agents and citizen observers. An armed group attacked the Makeni totalling centre on 15 August and took almost all the polling-station result forms. Witnesses told observers that alleged UPND cadres raided two of the three totalling centres in Kabwe on 16 August.

The ECZ said it would not publish polling-station results. Nobody here is asking for polling stations yet. The request for now is that the ECZ put on its own website the constituency results, presidential and parliamentary, that its officials read out on air. The Constitution allows seven days from the declaration for any challenge. The declaration was 18 August. Those seven days are the period in which the announced figures carry legal weight, and they are passing without the figures being published anywhere. If the official result is defensible, the announced numbers are its defence.

There is now a small industry of long words explaining why this request is unsophisticated. Give the ECZ time. Respect the process. Trust the institutions. Someone has proposed a longitudinal study. Of what, when the first data point is not published? And studied why: because the question is open, or because the answer is uncomfortable? I do not know, and I would rather not guess at anyone’s reasons. The request does not depend on them.

I have mates reminding me of 2021. I remember walking around the bank quietly encouraging people to vote for the man, even while my professional responsibility as an economist required neutrality in what I published. I believed Zambia needed change, and I believed he was the right change for the economy. He was.

You cannot be held responsible for everything a government later does because you supported it at an election. You are responsible for what you do once the evidence in front of you starts contradicting what you believed. You confront it, or you start finding reasons to excuse what you would have condemned from the other side. Across this continent we are better at protecting the image of our countries than at asking whether what is happening inside them is fair. We worry about what the world will think.

Sishuwa Sishuwa said many things I dismissed. At times I thought he was being alarmist. Today I find myself going back through what he said. Imagine that.

May this week never happen again. Unfortunately, we may need to pay more attention to the pattern that has presented itself. It is still 0 of 226.

Let that sink in.

AI-illustration: Four channels barred. One open. Zambia votes tomorrow. The Last Open Channel, Part 1: Before the Vote.

The Last Open Channel is a two-part series. Part 1 prices what the 13 August ballot tests, before the vote. Part 2, After the Count, reads the result through the same framework.

1. The Underpriced Ledger

Thursday’s contest is being fought principally inside a single ledger. The incumbent’s case is the dashboard: restructuring substantially complete, inflation back inside the target band, reserves rebuilt. The challengers’ case is the kitchen table: a price level that has not come down and employment that has not come through. Growth against diffusion is the headline argument.

The second ledger, the channel that runs from citizen to state, is present in the campaign on different terms. Brian Mundubile, the leading opposition candidate of the NRPUP-led Tonse Pamodzi alliance, has said the election must not be conducted under fear. His campaign has gone further, pledging to repeal the Public Order Act and the 2025 cyber laws within months of office, a pledge his running mate repeated in July. Gary Nkombo, formerly this government’s local government minister and elections chairman and now an independent parliamentary candidate, has described 2026 as a national struggle for freedom in the lineage of 1964 and 1991. The second ledger is priced low rather than ignored. The offer to reopen the channel is explicit, and its credibility is the open question. Mundubile served in the PF administration that enacted the predecessor cyber architecture, and his alliance supplied this week’s own evidence on democratic restraint. When the governance charge lands on the incumbent, the available answer is a trade-off claim: economic stability on one side, democratic norms on the other, one purchased with the other.

Two framing failures follow. The headline argument prices half the system. The trade-off claim asserts that one half must be surrendered to keep the other.

On Tuesday I ran a small, unscientific poll on X and on the Canary Compass WhatsApp channel. One question: which consideration should carry the greatest weight in choosing the president. On X, 244 responses: macro stability 39.3 per cent, jobs and living standards 27.5, democratic freedoms 18.0, the Constitution 15.2. On the channel, 21 responses: jobs 11, macro 9, the Constitution 1, freedoms 0. The four options reduce to three families: the dashboard, diffusion, and governance, since constitutional compliance and democratic freedom are one category. In both rooms the economic families outpolled governance, two to one on X and 20 to one on the channel.

Participation itself split along identity lines. The pseudonymous room answered in the hundreds. The identity-linked room, where a response attaches to a phone number and a name, answered in the tens and returned zero votes for democratic freedoms. The reticence is documented elsewhere. Freedom House’s 2025 Zambia assessment describes an environment of self-censorship built by arrests for online speech. The surveillance literature records the same withdrawal: measured disengagement from politically sensitive expression wherever the participant can be seen. A quieter mechanism runs through employment, where a public political position carries professional cost. A room that returns zero votes for freedoms may be measuring fear of the question as much as indifference to it.

The poll cannot show whether recognition lags erosion. It shows that, asked to choose one weight, both rooms placed economics ahead of governance. Graham and Svolik’s candidate-choice experiments in the American Political Science Review found that only a small fraction of voters prioritise democratic principles at the ballot, trading them against policy and partisan interests. The second ledger below documents what that ranking prices low. A ranking of that shape is the condition under which protections are cheapest to remove, whatever its cause.

2. The Framework: One Pipe, Two Directions

The framework is a transmission system with two directions. Downward, policy must pass through banks, prices, and credit to reach household budgets and farm gate prices. The state can also reach those endpoints directly, through administered prices and fiscal programmes. Direct delivery is bypass, not transmission: it works while financed, accumulates on a public balance sheet, and builds no transmission channel that survives it, whatever else it leaves behind. Upward, information must pass from citizens through speech, the press, and assembly to reach the state, including information the state does not want. Failure in the first channel produces improving aggregates without improving households. Failure in the second produces a state that receives only confirmation, which removes its capacity for correction.

One scope condition applies. The upward channel is not required for every downward outcome. Macro stabilisation is negotiated with creditors and the IMF and can be delivered without citizen feedback. Diffusion and integrity are different: knowing where households are failing requires demand signals, and detecting diversion requires functioning error reporting.

One evidentiary rule governs both ledgers. The incumbent is assessed on state power exercised. Challengers are assessed on record and conduct. The standard of evidence applied to both is identical. The test throughout: whether systems work, whether citizens advance, and whether dignity, economic and political, rises.

3. Ledger One: The Economy

The credit entries. The UPND inherited a sovereign in default, with arrears that had been insufficiently disclosed and inflation in the mid-twenties. Restructuring agreements covered about 94 per cent of the debt perimeter on the IMF’s May assessment. Inflation was 6.5 per cent in July, inside the target band. Reserves stood at USD6.4bn at end-April, or 4.4 months of import cover on the IMF’s measure; the Bank of Zambia’s own measure runs higher. Growth has averaged 4.5 per cent across 2022 to 2025, came in at 3.8 per cent in 2025 after revisions, and ran at 7.7 per cent in the first quarter of 2026.

The administered maize price rose from K150 to K347 per 50kg bag, a 131 per cent rise against a general price level up about 68 per cent. In real terms that is about 38 per cent higher. The improvement is a bypass, and the last two seasons flatter it. The administered price is one number set against a weather cycle. In the 2024 drought, private buyers paid K350 to K400 against FRA’s K330, and regional prices ran higher still. The floor sat below the market, and the agency struggled to buy in the year reserves mattered most. In the surplus seasons since, the market has priced below the floor, and the premium has been the support, at a cost the IMF now flags on FRA’s balance sheet as a quasi-fiscal risk. A fixed price fails in both directions across the cycle.

What the gate lacks is the infrastructure that would track the cycle: organised price formation, hedging instruments, and financing. The collateral routes exist in law: a movable property registry that is underused, and a warehouse receipts regime on the statute book since 2010 that has never financed a harvest at scale. A modernising act, No. 7 of 2026, followed this April; use at scale is untested. The IMF programme anchored the adjustment period and performed its function.

The debit entries are four, each a downward-transmission failure.

First, diffusion. The latest national poverty estimates, from the 2022 LCMS, record 78.8 per cent rural and 31.9 per cent urban poverty. Inflation measures the change in prices; households pay the accumulated level. From July 2021, the last month before this government took office, the general price level has risen about 68 per cent and the food price level almost 90 per cent. A 25kg bag of breakfast mealie meal averaged K141.77 then on ZamStats data; the July 2026 national average is K252.13, down from K275 a year earlier. Mean earnings of paid employees rose 53 per cent between the 2021 and 2024 Labour Force Surveys. A mean is pulled by its top earners and says little about the middle. More than half of paid employees earned K2,700 or less per month in 2024, and rural earnings grew materially more slowly than urban.

Electricity was the term’s most widely felt failure. On Afrobarometer’s 2024 fieldwork, published this May, only 28 per cent of households were connected to the grid, and among the connected, only 15 per cent said power worked most or all of the time. Combining connection and reliability, 4 per cent of Zambians had reliable grid supply. 83 per cent rated the government’s performance on electricity badly, double the 2022 disapproval. Households substituted privately: 47 per cent drew power from outside ZESCO, most of it rooftop solar.

The generation record runs the other way. Solar grew from 88MW in 2021 to 841MW this year on the ministry’s own count; Chisamba’s 200MW complex, completed in July, is the largest. The 750MW Kafue Gorge Lower, contracted under the PF in 2015, was fully commissioned in 2023. The IMF attributes part of its 2026 growth projection to recovering generation, and the fieldwork above predates the newest plants; the disapproval it records is what this electorate lived through. The critique concerns composition rather than effort. Solar was the fastest deployable answer to a hydro drought, and an emergency reaches for whatever deploys quickest. But the build layers variable supply on a hydro system whose failure mode is seasonal: daytime power added, evening peaks and the next drought still exposed. Firm alternatives ran on comparable clocks. Open-cycle gas delivers in one to two years, combined-cycle in two to three, and Mozambique’s gas-fired generation can reach Zambia through the Southern African Power Pool. Zambia has no domestic gas, so that route carries import exposure. Firm routes nonetheless ran on this term’s clock, and the build chose speed in one lane only.

The standard rejoinder is that diffusion is second-term work. The chosen instruments, free education with school meals, CDF at scale, and social cash transfers, are bypass instruments: direct fiscal delivery in place of transmitted income. The transfer rolls stand at 1.5 million households this year, about 8.4 million people, against 880,000 households at the 2021 handover. A drought emergency layer paid K400 monthly to new households for 12 months on World Bank financing and has closed. The Vice President cited the expansion on the trail this week. All are broad-coverage programmes with high measured approval. In the 2024 Afrobarometer survey, 84 per cent rated the government’s handling of education positively, the highest score on the continent, and 80 per cent reported a family member benefiting from free education. Coverage and approval measure reach. Lift would be measured by a national poverty survey, and none has measured this term: the 2022 LCMS remains the last word. A state that has not counted its poor since then cannot demonstrate diffusion, and the absence of measurement is itself a feedback entry.

Breadth requires less listening than targeting. Choosing which sectors and households to reach demands local information, and a narrowed upward channel carries less of it. These instruments face their first Zambian electoral test on Thursday. CDF at K40m per constituency and school meals at 4.6 million learners existed at nothing near this scale in 2021, and the transfer rolls have grown by 70 per cent. The cross-national precedents for this instrument class are unfavourable. Kenya’s first CDF election returned 77 of 183 sitting MPs, with the losers having run more projects than the winners, and Ghana’s school feeding incumbents lost a cost-of-living election in December 2024. Thursday supplies the first Zambian evidence on the instrument class.

Second, translation. The economic programme has been communicated to creditors, investors, and multilateral institutions, and not in terms accessible at household level. A programme its intended beneficiary cannot follow does not generate belief, and belief is a voting variable.

Third, monetary transmission. Genuine Kwacha private credit, defined as lending net of foreign currency loans and government exposure, stood at 7.4 per cent of GDP, computed on Bank of Zambia September 2025 data. Banks held government securities equal to roughly half of local currency deposits on the same basis. Under these conditions the policy rate mainly prices sovereign rollover rather than household and firm credit.

The counter-entry is legislative. SI 62 of 2025, the Banking and Financial Services (Capital Adequacy) Rules, took effect in January. The Banking and Financial Services Act, No. 9 of 2026, carries the close-out netting provisions a money market needs. Its own commencement clause leaves the operative date to a ministerial statutory instrument, and legal commentary dated 5 August still treated that order as pending.

The warehouse receipts modernisation passed as the Agricultural Credits and Warehouse Receipts Act, No. 7 of 2026, dated 8 April on the national law database. The Zambia Deposit Insurance Corporation Act, No. 11 of 2026, also dated 8 April, is enacted but not yet in force on the same database. Passage is not commencement, and commencement is not use: the 2010 regime and the banking act’s pending commencement are the proof. No administration in a generation has legislated transmission this substantively, and the framework is available to whichever government inherits it.

Fourth, the fiscal anchor. In July 2025 the government announced it would seek a 12-month extension of the IMF programme, an anchor through the election. The request submitted and approved was for three months. The government then dropped the extension route, completed the final review in January 2026, and entered the election year without a programme.

The IMF’s May staff statement records the outcome: fiscal pressures intensified by pre-election spending, a civil service wage adjustment, and agricultural subsidy overruns of about 1.3 per cent of GDP. The primary surplus is projected to fall from 3.1 per cent of GDP in 2025 to 1.1 per cent absent corrective measures. The outturn has already overtaken the projection. On the end-May numbers, the overall deficit reached 2.1 per cent of GDP, the full-year target consumed in five months, and the primary surplus stood at 0.2 per cent.

The sequence is a reversal of the government’s own stated request, at the point in the cycle where spending pressure was predictably highest. The authorities have reaffirmed commitment to a successor programme after the elections. That sharpens the finding: the anchor is wanted on both sides of the vote and absent only for it. The agricultural overruns in that finding are the bypass bill arriving. The support story and the fiscal pressure are the same policy viewed from opposite sides.

4. The Hinge: A Parabola

Corruption sits in both ledgers. The comparable index series begins in 2012, when the methodology changed; Zambia scored 37 that year. Across the PF decade the score slid to 33, with the trough reached under Lungu in 2020 and held through the handover. Under this government it rose to 37 in 2023 and 39 in 2024, the highest score Zambia has recorded on the index. It fell to 37 in 2025, the rank down seven places, the first decline in five years. The decline belongs to the PF decade; the recovery and the reversal belong to this government. The 2025 score equals the 2012 score: 14 years of movement net to zero, with the direction at the vote pointing down. Transparency International Zambia attributed the reversal to diversion of public funds through strategic business interests and political supporters ahead of these elections. That is the index custodian’s attribution, published in February.

The institutional record runs in two dated phases. July 2024: the ACC director general resigned amid kickback allegations. The board, whose chair, former Attorney General Musa Mwenye, had described the Commission’s management as pliant and answerable elsewhere, was dissolved within days. Commissioner O’Brien Kaaba turned whistleblower, alleging corruption within the state’s law offices. February 2026: the Commission, under new leadership, cleared Solicitor General Muchende, Copperbelt Minister Matambo, and Permanent Secretary Kawana. It announced that no cabinet minister was under active investigation, a declaration criticised by civil society and the former US ambassador. Kaaba’s dispute with the Solicitor General was settled out of court. The record therefore shows investigations reaching serving officials at the political tier and ending, in every high-profile instance, in clearance. One late case tests the pattern. In the campaign’s final fortnight, the outgoing Copperbelt police commissioner was charged with theft by public servant over copper cathodes worth USD448,596, alongside a magistrate and four others. The charge reaches the uniformed hierarchy; it does not reach the tier the clearances covered. The Lungu family faces prosecution. In the high-profile record, no serving member of the political executive has. The asymmetry, prosecution directed backwards and downwards, clearance directed inwards, is the operative finding.

The mechanism links the two ledgers. Sustained diversion becomes easier under two conditions: a delivery channel weak enough to feed on, and an error-reporting channel quiet enough to hide in. The corruption reversal and the legislative narrowing documented below occurred in the same period.

The precedent is 2006. Mwanawasa held the strongest macro position and the most credible anti-corruption record of any incumbent. He lost Lusaka and the Copperbelt to a campaign built substantially on the gap between a falling inflation rate and an accumulated price level. Household conditions decided the urban vote. This election runs under a threshold of 50 per cent plus one that no sitting president has comfortably cleared. Four indicators matter on Thursday night. The first is the urban vote, where diffusion is tested. The second is the threshold, where the Constitution binds. The third is the distance from the incumbent’s 59 per cent of 2021, which will show whether that result was a realignment or a receding wave. The fourth is the constituency vote in areas where CDF and school feeding penetrated deepest, where the breadth programmes face their first conversion test.

5. Ledger Two: The Upward Channel

April 2025: two statutes. The Cyber Security Act requires providers to install real-time interception capability and places the cyber security agency under the Office of the President. Interception requires a court order, obtainable ex parte as interception orders are. The concern is the breadth of the powers and the compulsory capability itself. The US embassy issued a surveillance advisory to its own citizens. The Cyber Crimes Act criminalises categories of online speech: harassment and humiliation, hate speech, and incitement of ethnic divisions, the last carrying life imprisonment. The terms are not defined. In a campaign analysed in regional blocs, the boundary of lawful comment on ethnicity sits with the prosecutor rather than the statute. The apparatus that classifies the speech answers to the office the speech would criticise. The UPND opposed the 2021 predecessor law from opposition; in office it re-enacted the architecture with shortened reporting lines.

The prosecution record across the term, under these and older statutes, is itemised and dated. May 2024: the PF secretary general was convicted of defaming the president and sentenced to 18 months with hard labour. Human Rights Watch recorded the conviction as coming despite the repeal of that offence in 2022. The repeal is a genuine credit. The conviction that followed it anyway is the entry beside it. 2024: journalist Thomas Zgambo was arrested twice and charged with sedition, a colonial-era offence, over posts about government transparency. January 2025: three people were arrested over statements about the president’s health. May 2025: former MP Munir Zulu was sentenced to 18 months with hard labour for a 2023 social media post. March 2026: two presidential candidates were charged under the cyber laws in one month. M’membe of the Socialist Party was arrested on 2 March over radio remarks on the delayed Lungu burial. Mundubile was detained at Kenneth Kaunda International Airport on 19 March and charged. May 2026: a 21-year-old blogger was arrested for sharing a video insulting the president. July 2026: a ZNBC journalist was arrested under the Cyber Crimes Act weeks before the vote, an arrest Amnesty International described as intimidation of journalists.

The Public Order Act, the colonial statute long used to restrict opposition assembly, remains in force. The current president was himself tear-gassed at home in 2017 and detained for four months on a treason charge later dropped. Parliament passed a replacement in May 2026. The President declined assent, citing inconsistency with judicial precedent. Whatever that reason’s merit, the statute remains available through this election.

December 2025: Bill 7. The Constitutional Court voided the original amendment process in June 2025 and directed a people-driven process led by an independent body. The government appointed its own technical committee, conducted nationwide consultations in the final quarter, revised the bill, and passed it with near-unanimous support, including PF members expelled by their party for voting in favour. The later process is fully on record, and the consultations were real. On the Court’s own language, neither satisfies the judgment. A government-appointed committee is a contested fit for the required independent body. Consultation conducted after the direction is fixed shapes the text, not the agenda. No further ruling has tested that reading. The same legislative period cut the voter register inspection window from 90 days to 14.

Below the statutes sits an unlegislated failure. The feedback reaching the state from its own side, its advisers, allies, and adjacent commentators, has become consistently favourable. Flattery and censorship produce the same output: approved information replacing accurate information.

The ledger carries credits. The Constitutional Court ruled against the state on Bill 7; the judgment stands, and its authors remain on the bench. The newspapers that condemned the Mongu police operation this week published freely in Lusaka. In the Zambia Election Research Network’s March survey, 51 per cent of respondents expected a completely free and fair election. Zambia is not an autocracy. Three days before the vote, on 10 August, Amnesty International published a briefing, based on May fieldwork, on what it called authoritarian practices eating away at human rights in Zambia. The documented finding is narrower: the upward channel has been restricted by statute, chilled by an itemised record of prosecution, and degraded by favourable filtering. Five years, under a government elected on a commitment to widen it.

6. The Closing Week

Democratic damage is recorded against whoever causes it. Both sides generated entries this week.

The documented record: Mundubile’s motorcade was blocked and stoned at Luampa, injuring a police chief inspector. Rallies at Mongu and Kalabo were dispersed with tear gas, and the Inspector General has ordered the operation reviewed. Campaign entourage members, including musicians and journalists, were detained at Mongu, 17 of them initially held without charge by the alliance’s count. Days before the vote, police raided the alliance’s media centre in Lusaka. More than 20 campaign staff were detained, by the alliance’s account without charge. Detaining performers, journalists, and the operators of a campaign’s communications in its final week is direct action against the carriers of the upward channel, whatever charges are eventually filed. The Barotse Royal Establishment, Western Province’s traditional authority, condemned the breach of order on its territory.

The reported and unresolved record sits between tiers, and it centres on two women: former ambassador Martha Lungu Mwitumwa, a campaign administrator, and Hannah Katontoka, who runs the candidate’s online platforms. The accounts conflict. Alliance-side reporting placed them among those detained in the raid. The family’s formal letter to the Inspector General describes an armed street seizure with live fire, and states that police stations confirmed neither woman was in custody, with no record of arrest. Videos of a street seizure circulated; they do not establish the men’s identity. On the campaign’s final day, the alliance reported members of its media and security teams still missing. A detention account and a no-custody confirmation cannot both stand, and the conflict between them is itself the entry. Detention becomes a process failure when its legal basis is undisclosed, custody goes unrecorded, or lawful time limits are breached. A seizure with no custody record is a category graver: the channel’s carriers do not face charges, they disappear from process altogether. On 10 August, Mwenye stated that Zambia now holds political detainees extra-judicially and appealed for their release or presentation to court. A published response disputed the characterisation, maintaining that fundamental freedoms remain intact. That the country’s former chief legal adviser and the government’s defenders are contesting the existence of extra-judicial political detention, in public, days before a vote, is itself an entry in this ledger. Resolving this record is a policing obligation that stands regardless of the result.

The further alleged record: Mundubile claimed cadres have operated in police uniforms and said Southern Province had been flooded with security personnel ahead of a planned visit. The alliance further alleged that funds raised to pay its polling agents were seized on 10 August, with police offering no comment by press time. Nine civil society organisations raised reports of SMS messages linking cash transfer payments to a vote for the president. These claims were unverified at the time of writing. The last allegation, resolved either way, bears directly on the results-chain demand.

The opposition’s entry: responding to ECZ guidance on ballot marking, Binwell Mpundu, the alliance’s presidential affairs chairman, stated on video that he was ready to die and ready to kill over the election. The video was deleted and attributed to AI generation and a hacked account. Mundubile publicly endorsed that explanation. Within a day, Mpundu confirmed the video was authentic and attributed the AI claim to his media team. Three distinct failures follow. The first is the threat itself, from a senior official of an alliance seeking state power. The second is the false AI attribution, which establishes that recorded evidence can be publicly disowned as machine-generated, degrading the country’s shared evidentiary baseline. The third is the candidate’s endorsement of a claim he could not verify, reversed by his own official within a day. The state restricts the upward channel by statute. The false attribution degrades it by precedent.

The count: three days out, the validity of ballot marks was still in legal dispute between the Commission and the alliance’s lawyers. The Commission has set 17 August as the expected announcement date, described by its own official as an administrative measure not prescribed by law. Its Chief Electoral Officer has committed to declaring the presidential result within 72 hours of the last station closing. Station results are announced and posted, and party agents receive signed forms. The narrower unresolved question, pressed by the opposition, is whether the progressive chain between station and national declaration remains continuously public. Zambian election credibility has historically rested on progressive announcement, verifiable by agents and observers as totals build. Disputed elections typically originate in the interval between polling and declaration. Claims made in that interval should be assessed on evidence, not affiliation.

7. The Cycle

The standard defence of the second ledger’s entries is that the same instruments were used against Hichilema. They were, and worse. The defence fails on its own structure: it measures the incumbent against his predecessor rather than against his commitments, which is the mechanism by which the instruments persist. State House defended this campaign’s violence as incomparable to the PF era. That is the same substitution, performed officially. The Public Order Act has served every administration since before independence. Over 60 years, the instruments that matter most have outlived every incumbent who condemned them. The repeal pledge now on offer has its own precedent: Hichilema promised repeal from opposition, delivered the defamation repeal, then re-enacted the cyber architecture. The stock behaves like the price level: the rate of abuse varies between administrations, and the stock does not fall. The alliance seeking to inherit that stock supplied this week’s own demonstration of how it handles restraint. The cycle requires no particular incumbent, only a successor willing to inherit.

The strongest argument against that successor is restoration risk, and it deserves full weight. The 2020 default, the under-disclosed arrears, and the institutional hollowing occurred under the PF. Mundubile and Makebi Zulu are products of that government. Markets would likely reprice Zambian risk within hours of a restoration. No single entry in either ledger carries more downside than an unmanaged restoration. Weighting it heavily is rational.

Three qualifications reduce it without removing it. First, the personnel are not confined to one side: the UPND has absorbed PF defectors into its ranks and candidate lists throughout the term. Second, a change of government is not a return to the past. Zambia has changed government three times: 1991 carried liberalisation forward, 2011 began the borrowing this ledger opens with, and 2021 produced the repair this ledger credits. Bleck and van de Walle’s comparative study of African electoral politics since 1990, with Zambia among its country cases, records continuity rather than rupture as the norm across alternations. Whoever wins inherits the constraint set rather than the predecessor’s choices: restructuring contracts with state-contingent clauses, a committed successor-programme path, and reserves that need markets willing to fund them. Third, structural reform does not prevent reversal; it raises its cost, slows it, and makes it visible. A repair that depends on its author retaining power indefinitely is not structure.

The ruin of a nation begins in the home of its people, and so does its salvation. The cycle ends when citizens stop measuring their side against the other side’s record.

8. The Last Open Channel

The headline argument prices half the system. The challenger offers to reopen the other half. His record and his alliance’s conduct price the offer. The trade-off claim fails on the scope condition: macro stabilisation never required the upward channel; diffusion and integrity did, and those are where the record fails. The corruption reversal and the narrowing of the channel occurred in the same period, and that co-occurrence is itself an entry.

Every other channel a citizen holds can be intermediated: posts interceptable under one statute, assemblies dispersible under another, journalism subject to investigation, submissions received after decisions are fixed. The ballot is the one signal a citizen originates without permission. Its carriage from station to declaration belongs to the Commission.

Vote with both ledgers open. Weigh the restructuring against the accumulated price level, and the transmission framework against the anchor requested and then dropped. Weigh the corruption reversal, the sequence of Bill 7, and the itemised record of detention against the credits the second ledger still carries. Hold one demand above the rest: the results chain, station by station, remains continuously public.

To whoever takes the oath in September: president after president inherited the instruments, condemned them, and retained them. The exit from the cycle is to dismantle an instrument while it retains its power. Dignity and order are the same system. Govern accordingly.

Sources

Afrobarometer, Dispatch No. 1178: Amid Electricity Crisis, Zambians Favour Ending State Monopoly, Investing in Solar and Wind Power (7 May 2026).

Amnesty International, Authoritarian Practices Eating Away at Human Rights in Zambia (London, 10 August 2026).

Bank of Zambia, monetary and credit data (Lusaka, September 2025).

Bleck, Jaimie and Nicolas van de Walle, Electoral Politics in Africa since 1990: Continuity in Change (Cambridge University Press).

Canary Compass, prior publications including the fiscal update of 31 July 2026, and author’s calculations.

CIPESA, analysis of the 2025 cyber laws (Kampala, May 2025).

Constitutional Court of Zambia, judgment on the constitutional amendment process (Lusaka, June 2025).

Electoral Commission of Zambia, public statements on results procedures (Lusaka, August 2026).

Freedom House, Freedom on the Net: Zambia (Washington DC, 2025).

Global Network Initiative, statement on Zambia’s cyber laws (2025).

Graham, Matthew and Milan Svolik, “Democracy in America? Partisanship, Polarization, and the Robustness of Support for Democracy in the United States,” American Political Science Review 114(2), 2020.

Human Rights Watch, reporting on the 2024 defamation conviction (2024).

International Monetary Fund, staff statement on Zambia (Washington DC, 14 May 2026).

International Monetary Fund, Sixth Review under the Extended Credit Facility, Country Report 26/21 (Washington DC, February 2026).

Ministry of Community Development and Social Services, social cash transfer caseload data (Lusaka).

Ministry of Energy, generation and access data (Lusaka, 2026).

Oxford Improving Institutions for Growth, briefing on Kenya’s Constituency Development Fund (Oxford).

Parliament of Zambia, legislative records (Lusaka, 2025 to 2026).

Press reporting: Bloomberg, JURIST, Lusaka Times, News Diggers, Open Zambia, Zambia Monitor, and Zambian Observer.

Transparency International Zambia, Corruption Perceptions Index commentary (Lusaka, February 2026).

World Bank, drought response financing under the Crisis Response Window (Washington DC, July 2024).

Zambia Statistics Agency, CPI releases and retail price series (Lusaka, July 2026).

Zambia Statistics Agency, Living Conditions Monitoring Survey 2022 and Labour Force Survey reports 2021 and 2024 (Lusaka).

ZambiaLII, Acts of Parliament Nos. 3 and 4 of 2025 and Nos. 5, 7, 9, and 11 of 2026 (Lusaka).

Disclaimer

This article does not constitute legal, financial, or investment advice. The author shares views for perspective and discussion only. Do not rely on them as a substitute for professional advice tailored to your specific circumstances. Always consult a qualified legal, financial, investment, or other professional adviser before making decisions based on this content. The analysis reflects proprietary research undertaken by Canary Compass and the author.

Canary Compass and the author accept no liability for actions taken or not taken based on the information in this article.

The views expressed in this article represent the author’s independent professional analysis and do not constitute an endorsement of any individual, institution, or position. Canary Compass and the author accept no responsibility for how this content is interpreted, excerpted, or recontextualised by third parties not involved in its production and publication. Reproducing any portion of this work in isolation, or in combination with other material, in a manner that misrepresents the author’s original meaning constitutes a distortion of the published record.

The author may hold positions in financial instruments, currencies, or assets discussed or referenced in this publication. Such positions do not constitute a recommendation to buy or sell.

All views, projections, and forecasts reflect the author’s assessment at the time of writing. Data sourced from third parties is believed to be reliable but has not been independently verified. Past performance does not indicate future results.

All content published by Canary Compass is the intellectual property of the author. Reproduction, adaptation, or redistribution, in whole or in part, requires written permission.

About the Author

Dean N. Onyambu is the Founder and Chief Strategist of Canary Compass, a financial research publication focused on African monetary architecture and financial sovereignty. He brings 18 years of experience across trading, fund leadership, and economic policy, with senior roles at Standard Bank, First Capital Bank, and Opportunik Global Fund.

Read and subscribe at www.canarycompass.com.

The Canary Compass Channel is available on @CanaryCompassWhatsApp for economic and financial market updates on the go.

For more insights from Dean, you can follow him on LinkedIn @DeanNOnyambu or X @InfinitelyDean.

AI-illustration: Reopening the Ledger

The football debate this week took me somewhere larger than football. The immediate trigger was FIFA’s proposal to bring private investment into a new commercial structure while using the proceeds to expand development funding across its 211 member associations. As the argument widened, I found myself back at one member association, one vote, and at a harder question: when may an economic difference legitimately become a formal institutional difference?

Institutions have always answered that question differently. The United Nations General Assembly gives each member state one vote. The United Nations Security Council privileges five permanent members. The International Monetary Fund (IMF) substantially weights voting power through quotas. These are different settlements of power, built for different purposes under different political conditions. None is correct merely because it already exists. Equally, weighted voting is not wrong simply because it is unequal.

The burden sits on the design itself.

Any differentiated formal authority, inherited or newly proposed, should have to establish why the variable being weighted is relevant to the institution and why differentiated authority is necessary rather than merely convenient. The degree of differentiation must also be proportionate to the responsibility or contribution being recognised.

It should also ask whether the same legitimate objective can be achieved while changing fewer fundamental rights. That is what I mean by a less distorting mechanism. If the problem concerns commercial decisions, the answer need not automatically be a permanently heavier constitutional vote. Stronger commercial representation, a different voting rule for specified matters, a revenue arrangement or a time-limited accommodation may solve the narrower problem without rewriting the entire franchise.

Any privilege justified by a particular condition should also remain genuinely reviewable when that condition changes. Reviewability cannot mean merely that a rule may theoretically be reconsidered. There has to be a workable route for testing whether the justification still holds, including access to the evidence on which the weighting rests.

That sounds cleaner than it is because institutional purpose is often part of the dispute. Is an institution principally a forum of members, a financial pool, a regulator, a development vehicle, a commercial enterprise, or some combination of them? There is no neutral observer who can simply announce the answer. The starting point has to be the institution’s governing bargain, its constitutive documents and its settled functions. Where members want to redefine that bargain, the change has to pass through the authorised constitutional process. The faction that benefits from a particular interpretation cannot simply declare that interpretation to be the institution’s new purpose.

Power complicates this further. Institutions are often settlements of power before they are expressions of principle. A member may genuinely possess capital, market access, participation or cooperation without which an institution would be materially weaker. That creates leverage. It may even require accommodation.

But indispensability is a fact about bargaining power, not a permanent title to rule.

The dependency must therefore be current rather than inherited by assumption. Any accommodation should be no larger than the dependency requires, and narrower alternatives should be considered. If the dependency disappears, a privilege created to manage it should not survive merely because everyone has grown accustomed to it.

The IMF is worth actually running through this framework rather than waving at it.

Its quota formula gives GDP the largest weight, alongside openness, economic variability and reserves. Quotas help determine financial contributions, voting power, access to Fund resources and shares in general Special Drawing Rights (SDR) allocations. There is therefore a real functional relationship between economic differences among members and parts of the Fund’s architecture. That gets the argument through relevance. It does not finish the argument.

Why should those variables affect voting power to the degree that they do? Does the resulting allocation remain proportionate as the world economy changes? Does the mechanism adjust quickly enough when the economic relationships it is supposed to represent move?

On the Fund’s own evidence, the principle survives the relevance test, but the present distribution does not fully pass the adjustment test. Representation gaps remain, and the membership itself acknowledges that quota shares need to better reflect changes in the world economy.

That does not make quota weighting illegitimate. It shows why a functionally defensible principle cannot give its current distribution permanent immunity from review.

Nor do I object to economic power itself. Market access is power. Capital is power. Technology, tariffs, sanctions, aid and control over financial infrastructure are power. America uses them. So do Europe and China. African states should use credible leverage wherever they possess it too. I have supported America’s willingness to exercise raw economic power where I believe it creates strategic space for Africa. I cannot then object to another actor merely because it possesses leverage or uses it to press for institutional change.

The harder issue begins when an economic variable is proposed as the basis for a formal right.

Before anyone can say that contribution should purchase authority, they have to define the contribution they are trying to measure and explain why it matters. If the relevant contribution is capital at risk, measure capital at risk. If it is commercial contribution, define whether that means revenue, investment, audience, intellectual property, talent development or some combination. A balance sheet cannot become a constitution while leaving the accounting concept undefined.

Once present economic power is invoked to justify formal authority, the provenance of that power becomes a legitimate part of the inquiry.

History may therefore matter to the exercise, but it receives no evidentiary shortcut. If historical extraction materially helped create the capital, infrastructure, ownership position or productive advantage now being counted, that may change how the present structure should be understood. The claim still has to demonstrate a material causal connection, with attribution strong enough to survive scrutiny. Later investment, domestic policy, technology, institutional quality and other intervening causes remain part of the account.

Historical injury receives no evidentiary shortcut either. If the causal chain is weak, the effect immaterial or later factors overwhelmingly explanatory, the claim should lose.

History may change what we understand today’s economic structure to be. It may reveal that a proposed metric misattributes contribution and therefore influence institutional redesign. What it cannot do is mechanically convert yesterday’s injury into today’s allocation of formal authority. Yesterday’s extraction is not a voting formula any more than today’s wealth is one.

That inquiry does not settle the reparations question. Reparations remain a separate corrective claim. If a historical wrong, transfer and causal chain can be established, pursue it on its evidence and merits. A successful claim can alter resources and bargaining positions. It can even provide capital for development.

A historical invoice does not build a refinery. An honoured one may help finance it.

Finance can buy equipment, training and technology. It can help build institutions and productive capacity. But the transfer itself does not supply engineering discipline, reliable energy, competent management, sound procurement or the operating capability required to make the refinery work. A valid historical claim can become an input into development. It is not a substitute for development strategy.

Resource sovereignty and resource nationalism sit further downstream. A state may conclude that too much value from its resources has historically been, or continues to be, captured elsewhere. It may then change the terms of future extraction, processing, ownership or taxation. At the extreme sits nationalisation. History may strengthen the case for reconsidering an ownership structure, but every policy must still survive its own tests of purpose, legality, compensation where applicable, capacity, capital requirements and consequences.

The same discipline applies to counter-escalation. A response can be strategically rational without necessarily being wise. It needs a referent, but that referent need not be a single event. It may be a rule change, an ongoing institutional arrangement or a cumulative economic structure. The response should address the relationship being challenged rather than attach an existing policy preference to history in general and call it retaliation.

Deterrence requires more than a valid claim. Evidence may establish the claim. Capability determines whether anyone has to care.

Which brings me back to football.

If commercial contribution eventually becomes the proposed basis for reweighting formal voting rights, “we make more money” cannot settle the argument. The proposal would first have to define what FIFA is being asked to become and pass through the constitutional process authorised to make that choice. It would then have to define the contribution being measured, show why it requires differentiated formal authority and demonstrate that the proposed weighting is proportionate.

It would also have to answer the narrower question. If the dispute concerns commercial influence, why must the solution be a permanently heavier Congress vote? Could the problem instead be solved through stronger commercial representation or a different rule confined to specified financial decisions?

The other side carries a burden too. If it answers with history, extraction or unequal value capture, those claims must survive the same demands of causation, materiality and attribution. No one gets to skip the ledger because their preferred claim is morally attractive.

If you price political voice by today’s wealth, expect others to price today’s wealth by history, and expect both prices to be checked against the same ledger.

The ledger may reopen. What follows still has to be designed.

Have a peaceful weekend.