Harvest record — the Navier–Stokes blowup-formalization chat

Filed as a record so the harvest survives, not only its outputs. Kept verbatim below the line; this block is the outcome.

Outcome, 2026-09-13. §1 filed as cases/CASE-21-swirl-annulus.md (name shortened from the digest's suggestion; frontmatter supplied at filing). §2 amended into lab/runs/027-axisymmetric-swirl.md — predecessor result, the sweep as three runs with run1 the discriminator, settles: CASE-21; blocked_on left standing, the run is still draft. §3 filed as probes/P-005-singular-limits-commute.md with relation: undetermined to CASE-07 — the mechanism test is the probe's own job, not a call made at filing. §4 filed in OPS.md as a deductive check, with the commutator read as the uncertainty principle stated about an algebra rather than a measurement. §5 skipped deliberately. CASE-07 is referenced but not in this repo's cases/; the reference stands.

---

Harvest — "AI formalization of Navier-Stokes blowup proofs"

Retroactive harvest, 2026-09-13, from a chat whose only artifact reaching the repo was axiswirl.py. Source: chat c3333105-7461-4fbc-ad90-549cba7566c0. Items below are quoted or closely paraphrased from the transcript, not from a summary — but confirm each before filing, since a retroactive harvest cannot tell your decisions from my suggestions as reliably as a live one.

---

1. A complete case, never filed — file as written

It exists in full house format: presenting object, observations, friction with two accounts failing in opposite directions, question for the panel, data provenance. Nothing needs writing.

Case: the unforced swirl annulus that builds the collapse geometry and stops short.

Axisymmetric NS with swirl, z-periodic, unforced, ν = 2×10⁻³, Rayleigh-unstable exterior. Over 14 time units: fluctuation energy grows e¹³ and saturates; annular Reynolds stress ⟨u_r′u_θ′⟩ positive (+0.64) — outward angular-momentum flux; its divergence exceeds viscous diffusion (0.044 vs 0.038); saturated state is a single meridional cell; the local swirl maximum decouples from the z-mean, sits 7.8% above it, and stops decaying; its radius does not move, r ≈ 0.5.

Friction. Account A — a Rayleigh-unstable profile mixes itself flat, nothing collapses. But the instability produces the collapse geometry with the right stress sign, and the core stops decaying. Account B — the instability sustains the collapse, as in the forced construction. But the cell delivers inflow strain ≈ 1, fixed by cell size, where collapse requires 4/τ growing without bound. The stress direction is pinned by energetics; the magnitude is unexplained.

Question for the panel. What sets the magnitude of the annular Reynolds stress relative to the collapse requirement in an unforced flow — Reynolds number alone (a threshold Re_θ above which cell strain outruns core diffusion), non-axisymmetric modes the axisymmetric run cannot represent, domain geometry (cell scale set by box rather than core), or never (mixing always wins, h = 0 is forced)?

2. Run 027 is missing its own predecessor — amend the run file

027 exists to push past a result its file does not mention: the local swirl maximum arrests but does not contract at Re ≈ 1500. Writing 027's prediction without that is writing it blind to the thing it predicts against.

And the sizing is already specified, which 027 also lacks. The queued sweep from the chat: --hp 0.5 --nu 3e-4 (stronger instability, higher Re), then --ra 0.8 (narrower annulus, closer to the core), each with its own tag. The case states the discriminator outright: a satisfactory answer should predict whether the radius of the local swirl maximum contracts at ν = 5×10⁻⁴, box length 12, annulus on r ∈ [0.35, 1.2].

That is 027's falsifier, already written, in a chat.

3. A probe — and it connects to an existing case

The "two faces" idea was resolved in-chat, not left open: pressure nonlocality and dissipation are the same angular projector at different points on the Mellin line, and the Leray projection is what bulk viscosity becomes when the radial degree drops from 2 to 0 in the infinite-sound-speed limit.

What it generated is open, and was flagged as the place to push:

> Do the two singular limits commute? c → ∞ and ν → 0 act on different sectors. Onsager's > anomalous dissipation is a transverse-sector statement about ν → 0 at fixed c = ∞. Does > approaching the Euler singularity along a path with c large but finite behave differently — > is finite sound speed a more natural regularization of NS than hyperviscosity, since it > keeps both faces at degree 2 with nothing nonlocal to fight?

This is CASE-07 (closures don't commute) in a new subject. Worth filing as a probe with an explicit cross-reference — and worth checking whether it is the same non-commutation or merely the same shape, which is the mechanism test.

4. A method lesson — deductive, with the argument in one line

The framing was declared blind to transport, and the reason was given rather than observed:

> Fourier multipliers are exactly the operators commuting with translation — functions of > momentum p alone. Transport u·∇ is a function of x and p. Since [x, p] = i, nothing > built from p alone can express it.

Generalised: the operator algebra you choose determines what you cannot see, and you can say in advance what it cannot see. Not an inductive pattern — a consequence of which group the algebra is the commutant of. Belongs in the deductive tier.

It also came with its own scope statement, unprompted: "this program characterizes the linear and quadratic operator content precisely and leaves the actual enemy exactly where it was."

5. A corrections ledger, and a preference

Four corrections were logged as the conversation ran: curvature → connection (flat); backward paths equilibrate inward → only subcritical, supercritical pushes outward; pulses buy h → h is exterior-inherited; Lions 5/4 misattributed to Tao.

The practice itself became a running feature — worth recording as a preference, since it is the same discipline as this project's corrections-on-record and arrived independently.

---

Filing summary