The missing plan.

The most-documented private funder in the OECD's books spent 26 August 2026 saying the quiet part in every register he has.

Bill Gates published an essay that day about the age his own industry is building. One sentence in it does the work of the whole: 'There is no plan to ease the entry into the AI era.'

He did not say it once. To the New York Times, in an interview the same week, he said the industry knows and downplays the danger: 'In private, people who understand how good this stuff is, and how much better it's getting, they're very worried.' Asked about the voluntary safety measures industry and government were preparing, his answer was two sentences: 'Self-regulation on the most dangerous tool ever invented? No, thanks!'

And to be fair to him, because this page quotes the strongest form of his position, the essay does not stop at alarm. Its closing section is headed 'The world needs a plan', and he proposes institutions to make one: national bodies that can set priorities, and an international organisation modelled on nuclear inspection.

So a funder whose giving the OECD profiles line by line believes three things at once. A transformation measured in trillions is running without a plan. Self-regulation is not a plan. And the fix is institutions that verify.

Every quote in this section is a design statement, published by the system about itself. None of it is scandal, and that is the point.

The OECD's current profile of the Gates Foundation reports that it screens 99.5 per cent of its bilateral allocable aid activities against the gender equality policy marker, and committed USD 2.2 billion a year to gender equality in 2023 and 2024. Screened is the system's own word.

The system doing the screening describes itself, in its binding text, in one sentence: 'Policy marker data are descriptive rather than quantitative.' The methodology, the directives explain, is purpose-based: activities are marked by stated objectives, 'not primarily in relation to their relevance or outcomes'.

The newest marker's handbook, for disability inclusion, says it plainest: 'The marker cannot and does not intend to measure the outcome or impact of a project.' And on checking: 'The Secretariat is not able to conduct in-depth checks for each reporter.'

This is not one marker's quirk. Eleven markers ride one reporting system, one field list, and no field in it records what any of the money did. Gender is the largest at USD 65 billion a year of marked commitments, climate follows, disability is the newest. Self-declaration, spot checks impossible, outcomes out of scope by design. There is a name for that arrangement in the Gates interviews: self-regulation.

We know what happens when someone calls the bluff, because someone did, and we hold the documents.

In October 2021 the director of the World Food Programme said two per cent of Elon Musk's wealth could help solve world hunger. Musk answered with a condition that should be famous: describe exactly how this solves it, 'But it must be open source accounting. So the public sees precisely how the money is spent.'

Nine days later, WFP published the plan: USD 6.6 billion, 42 million people on the edge of famine, 43 countries, cost lines given. We hold the document in this project's archive.

Then the record stops. Musk's Tesla shares, about USD 5.7 billion of them, were gifted that November to his own foundation. WFP's director later confirmed on the record that his organisation received none of it.

No villain is required to explain this. The demand was verification. The plan was intent. Between them stood no instrument that could turn one into the other, and where there is no instrument, there is no deal. That is what a missing plan costs, measured once, in public, in dollars.

Gates proposed it for AI without noticing aid already needs it: an institution that verifies. That is buildable, because we are building it.

A plan, in aid as in AI, is not a promise and not a marker. It is machinery: a register where every claim carries a source anyone can open. An archive that holds the source, so a dead link cannot delete the record. Numbers pulled from the primary systems by tools anyone can run, so the figures reproduce instead of being believed.

The first version of that machinery runs on this site. It covers all eleven markers, every topic they carry, both sexes, and every country the register names. It started with one question about one marker, because that is where we happened to begin counting, and the same field list that cannot answer it cannot answer for climate, disability, nutrition or anything else it counts. The machinery is for all of it, and for anyone.

The essay is right. There is no plan. There could be.