Other material · A dividing-plane barrier in the OpenAI forced Navier-Stokes blow-up construction

Index of the work on the OpenAI manuscript, September 30, 2026

The opening page of the material on the OpenAI forced Navier-Stokes blow-up manuscript, written on September 30, 2026, the day the line of work behind this note's question began; the owner of Hypnos, the research harness this site describes, ran that work with frontier-model sessions. It lists the primary sources with their retrieval dates and SHA-256 fingerprints, what each file holds, the public context and a one-paragraph summary of what stands. It calls the files by their private names (LEDGER.md, ledger.json, EXPLANATION-<section>.md), published here as ledger.md, ledger-v1.0.json and explanation-*.md, names a folder of digests that is not published, uses the project's internal stage names, and carries addresses of other sites, shown here as text.

Written by: Claude Fable 5.1 (Anthropic)
Size: 5,634 bytes
SHA-256: 6a55357dcb3d4f12226823f01ec3057280ec70e199e1550f9f78667611515051

The note's page Every file published with it This file on GitHub

The 2026 forced Navier-Stokes blow-up: sources, ledger, and the pull-back record

Owner-directed 2026-09-30 (registration: docs/DESIGN-BLOWUP-PULLBACK.md). This folder holds the record stage I files from and stage II grades against. Times in this folder are UTC unless marked EDT.

Primary sources (the ledger is built from these and nothing else)

Source	Retrieved	Identity
OpenAI, "Finite Time Blowup for Navier-Stokes", 166 pp.	2026-09-30 19:46Z, `https://cdn.openai.com/pdf/32d9f210-8b73-45e0-91bc-82a30aef8a9a/navier-stokes.pdf`	sha256 `0e779481c4da40bd28d1e642e1d8ca57447d129610df28dfa5a11e9af8ae228f`, 2,959,204 bytes
OpenAI, "Finite Time Blowup for the Euler Equation", 57 pp.	2026-09-30 19:47Z, `https://cdn.openai.com/pdf/315b36cd-ec98-4023-8342-93345194ece1/euler.pdf`	sha256 `a0c234518e6c489e16996805023eb2e75c00b7c03455f7a3a5be2c124954bfdd`, 535,142 bytes
github.com/openai/NavierStokesAndEuler (Lean 4.34.0-rc2 over Mathlib)	2026-09-30	2,669 files; 816 modules under `NavierStokes/`, 1,839 under `Euler/`; Comparator targets `NavierStokes.Comparator.navier_stokes_breakdown_R3`, `navier_stokes_breakdown_periodic`; permitted axioms propext, Quot.sound, Classical.choice; last push 2026-09-10
Buckmaster, statement, 4 pp.	2026-09-30, `https://cims.nyu.edu/~tristanb/statement.pdf`	sha256 `8d7723941bcda2fa55c1e74faa6298e04c706d17ff8abd2ad01878039c621f9d`
The literature ledger's arXiv ids	2026-09-30, each verified at `arxiv.org/abs/<id>` by the literature pass	recorded per entry in `LEDGER.md`

The PDFs and their extracted text are not committed (they are OpenAI's documents); the hashes above identify the exact bytes every page and equation reference in the ledger points at.

Files

LEDGER.md: the mechanism ledger, human form. One block per move of the proof (sections 4 to 10, appendices A to C), per move of the Euler paper, per lineage result, per wall, and the endpoint. Each block: statement, obligation, mechanism, antecedent, cost, the backward question, a checkable computation or "none: pure estimate", pages and equations.
ledger.json: the same, machine form, the input of scripts/file_ns_blowup_corpus.py (stage I) and scripts/blowup_pullback.py (stage II). Its SHA-256 is recorded in every manifest that consumes it.
digests/: the section digests the ledger was assembled from, as written by the digesting agents (one per section), kept verbatim as the working record.
EXPLANATION-<section>.md: the cortex half of stage II, one motivated explanation per section, every claim tagged with the page it rests on; assisted research, ungraded, never autonomous production.

Context (paraphrased; not sources of any ledger entry)

The video that prompted the ask: Matt Parker, "Has a Millennium Prize Problem been solved by AI?", Stand-up Maths, 2026-09-21 (youtube.com/watch?v=c79PsbkHw4A). Segments: the equations and the Clay alternatives (01:38 to 07:29); the result, the schematic and the construction read from the paper, including the residual trick and the pulses (07:29 to 13:40); the priority dispute and the cost figures (13:40 to 16:11); why mathematicians object, the lost journey, the squaring-the-circle and quintic analogies (16:11 to 18:56); pure against applied, and what happens next (18:56 to 22:10).
OpenAI's announcement (openai.com/index/navier-stokes-solution/, 2026-09-08): about 10,000 concurrent agents, 88 hours, on the order of 130 billion output tokens; the Euler problem set first; the Lean translation 17 further hours; the prize not claimed.
Clay Mathematics Institute (claymath.org/news/navier-stokes-announcement/, 2026-09-11): the problem "has apparently been settled"; the review is deliberately unhurried; recognition follows publication and community vetting.
Tao's blog, September 2026: "After Math" (De Toffoli and Duede, 2026-09-12), "Happy, those able to know the causes of things" (Guillen, 2026-09-13), "If math is more than proof, we need to better celebrate the rest of it" (Sanderson, 2026-09-18), and Tao's own "Headlines and inside stories" (2026-09-23). Gamburd, "The Siren Call of Silicon Leviathan", arXiv 2609.28591 (2026-09-20).
Press: CNBC 2026-09-09; NPR 2026-09-22 ("humanity learned nothing (so far)"); Tufts Daily 2026-09 (Karniadakis on the 70 nm scale where the continuum assumption fails for air; Fefferman on Cordoba and Martinez-Zoroa).
The literature map at navier.microprediction.org (pages: equations, residual, scaling, core, fluctuating, formal, status, map, timeline).
Buckmaster's statement: the three results (IPM, Boussinesq, 3D Euler with smooth forcing; August 15; Lean August 22), the credit to Cordoba and Martinez-Zoroa, and the account of the September 3 to 8 exchange with OpenAI.

What stands and what does not, in one paragraph

Alternatives (C) and (D) of Fefferman's statement are claimed with a Lean certificate and a 166-page manuscript that no human had read in full as of 2026-09-20. Alternatives (A) and (B), the unforced problem, are untouched. The construction is a forced, exactly balanced, non-attracting flow; three post-publication papers constrain it (analytic forcing near the singular point forces regularity; the core geometry does not transfer to the similarity variables as a boundary value problem; blow-up-producing forces are dense exactly below H^{1/2}), and one names a target for the unforced search (the positive defect problem). The stance rule of docs/DESIGN.md section 1 applies to every sentence in this folder.