The development finance architecture measures itself in the lender's currency and the lender's year

The international development finance architecture keeps two ledgers. One records commitments, disbursements, the private money that closed alongside them and the terms on which anything left a balance sheet. Every entry in it can be dated and audited. The other ledger records what the money was for: a grid that stays up, a port that clears cargo, a health system that holds through a bad year. The first ledger closes each December. The second does not close at all. Most of what is wrong with the incentives in this system follows from the fact that only one of the two can be reported on time.

Where this is written from

I write from the editor's desk of this brief, working from federal statistical series and public documents. The position has one advantage worth naming. Prices are public. The rate at which the United States government borrows for ten years, and the value of the dollar against a basket of trading partners, are published on a fixed schedule, carry a source line, and can be checked by anyone who disputes what has been done with them here. Nothing in the two sections that follow depends on private information.

The same desk has a larger disadvantage, and it bounds this piece. No data on any development finance institution is held here. Not a portfolio, not a set of loan terms, not a capital position, not a result in any country. I have sat on no credit committee and read no country strategy that was not published. So this is not a finding about anyone's conduct and could not be. It is an argument about the shape of a system, built from rules the institutions publish about themselves and from two prices anyone can pull. Where the argument needed a fact this desk does not hold, the sentence has been left unwritten rather than filled in.

A concession is priced against something

Concessional lending is defined by a comparison. A loan is concessional because its terms sit below what the borrower would otherwise pay, and the benchmark for that comparison is a market rate of some kind. Which rate any given institution uses is set in its own rulebook and is not held at this desk. What can be shown here is that market rates moved, and by how much. The ten-year Treasury constant maturity rate averaged 0.89 percent in 2020 and 4.29 percent in 2025, a rise of 3.40 percentage points across the six annual observations this publication holds. It rose at every step. That series is the price of United States government credit and nothing else, and it is offered as the public marker of the move rather than as anyone's discount rate.

How it rose matters more than that it rose. The single-year steps run 0.56 points into 2021, then 1.50, then 1.01, then 0.25, then 0.08. The last two are the smallest of the five. Whether the series has finished moving is not something six annual observations can settle, but a rate that jumped and then changed by 0.08 in a year is not behaving like a disturbance passing through a rulebook written in a different decade. It reads as the ground that rulebook now stands on. A threshold or concessionality test calibrated when market rates sat near zero is measuring against a different level today, and a test does not announce when the level beneath it has moved.

The unit of account belongs to the lender

The second misalignment is older and gets less attention. Obligations are written in the lender's currency. The broad dollar index, with January 2006 set at 100, averaged 113.1 in 2021, the lowest reading in the window this publication holds, and 122.7 in 2025. That is 9.6 index points, or 8.5 percent of the 2021 figure. The index is a trade-weighted basket and is not any borrower's exchange rate; what it marks is the direction of the unit the obligation is written in, not the size of the gap for any one country. A borrower whose revenues are collected in its own currency, from its own tax base, carries whatever that gap turns out to be. The architecture prices credit risk with great care and keeps no standing instrument for this one. The absence is not an oversight that nobody has noticed. It is a decision about who holds a risk, taken by default and renewed every time a facility is documented.

What can be counted on the day it closes

Between those two observations sits the mechanism. An institution reports on an annual cycle to shareholders who sit in finance ministries and are themselves appraised annually. What can go into that report is what is true on signing day: the size of the facility, the money that moved alongside it, the terms agreed. What the institution says it exists for is true, if it is true at all, long afterwards, in a place where nobody is counting any more. Staff are promoted against the first horizon. Boards are briefed against the first horizon. The second horizon has no owner inside the building. A system does not need bad faith to drift towards the number it can produce on time.

The objection with real force

There is a serious answer to all of this and it deserves stating at full strength. An institution that cannot report a number cannot be funded. Appropriations are annual, legislatures are sceptical, and a body that meets every question about its performance by explaining that the answer will arrive in a decade will lose its budget to one that does not. The countable figure is not vanity. It is the price of remaining an institution, and anyone proposing to remove it is proposing to stop the work. The objection holds, and it does not dispose of the problem. The requirement is that something be countable, not that the countable thing be the disbursement. A figure whose period matches the claim would satisfy the same constraint. A portfolio still performing long after the facility closed is countable. A cost of capital achieved against the benchmark as it stood on the day is countable. A currency mismatch carried, and named, is countable. Each of these is harder to produce and arrives late, which is the whole of the reason none of them is the headline figure.

What I would change

Three things, and then the limit. Publish the benchmark used in every concessionality test, with the date it was set, so that a test calibrated in 2020 cannot be quoted in 2026 as though the comparison had stood still. Report the currency of the obligation and the currency of the borrower's revenue beside each other on the face of the document, whether or not anything is done about the gap. Separate the signing-day report from the outcome report, so that the two stop being read as one document by people who have time only for the first. None of this follows from a dataset, because no dataset here reaches inside these institutions. A reader entitled to check the arithmetic on two Federal Reserve series is not thereby entitled to take my word on the third thing. The rate and the index are facts. What has been built on them is an argument, and it should be read at that weight.