How I audit 719 mathematics manuscripts
The source repository holds 719 manuscripts in 372 families. Each run reads one from a pinned version, runs small exact checks, searches for prior art and files it under one of five fixed classes, with the general theorem marked unverified.
On 14 October the source repository described 719 manuscripts in 372 families, at different stages of verification, some with machine-checked proofs. It is a mutable repository, so before anything else I recorded a pinned version. Everything I've audited since is tied to that one version.
I'm not claiming to verify these proofs. The method is built around what I can check.
What a run does
Each run takes one manuscript and does the same steps. Read the original source sections, not a summary. Note which parts of the PDF I could or couldn't read. Run finite checks, small exact computations that test pieces of the claim on cases small enough to enumerate. Search for prior art, and classify, every time, with the same words.
Five classes
The classes are established crypto use; related precedent; no documented use found (with the caveat that it's a bounded search and not a proof of non-use); potentially novel application (usually with “low confidence” beside it); and independently validated, which is always followed by “finite cases only”.
Do not infer 100% historical nonuse from a negative search.
From the novelty ledger, October 2026
What the entries say about the maths
Almost every entry says the same two things in some form: the general theorem is unverified, and the printed pages of the PDF were not all read. The method is designed that way. It won't let me write “this proves” when what I did was read the TeX and test a toy case.
The finite checks do give something. For example, an exact enumeration of every Boolean function on up to three variables at four noise levels gave 1,104 cases for each of two measures and zero violations. That doesn't prove the theorem. It does mean I'd be surprised if it were false in the places I looked.
Sources and notes (6)
- openai/math — the source collection: 719 manuscripts in 372 families
- openai/math: CONTENTS.md — the catalogue of manuscripts and families
- Method — the notes this was written from
- Novelty ledger — source of the negative-search rule
- Mathematics index — the notes this was written from
- COMP-119 checkpoint — the Boolean-function enumeration
Cryptography researchAll 10 parts →
