Will Cook
Public work
I build a private system for coding agents to work in, and the tools to inspect what they did. The public parts are the ones a stranger can run, check, read or watch.
Read it with an AI
The whole public record is also published as one packet: a single JSON file you can give to an assistant instead of browsing the pages yourself. It contains every public page and record, each component claim with the limit on that claim, and instructions to cite exact pages.
Download the packet .json save it, then drag the file into ChatGPT, Claude, Gemini or Grok. There is a smaller digest, also JSON, if your assistant refuses a file that size. The reviewer brief is shorter still and carries only the decision layer: what a reader in your position can and cannot conclude from the public record.
The assistant reads and quotes the public files. It cannot run any of the checks. If it tells you it ran one, it is wrong.
Why it is here
I am 22, an economics undergraduate at Bristol. I built this alone over about a year and paid for it myself, and AI was the only tool I used.
I designed the system for coding agents to work in, on the assumption that it gets more useful as the models improve. I have not shown that; it is still an assumption. Most of the mathematics is written in Lean, which checks every step, so you do not have to take my word for those claims. The Erdős problems the work covers are open, and none of it solves one.
Nobody independent has checked the mathematics yet, and nobody outside has seen the whole system. If you can evaluate work like this, the most useful thing is bounded: take one public claim, follow it to its source and its evidence, and tell me where the wording goes further than the record does. If you are in a position to look at all of it, I would like to show you. I am also looking for work or funding; a fellowship or something like it would let me do this properly instead of alongside a degree.