A regime-2 witness now holds at two Grashof points.

The Load-Bearing Statement

Two-regime, dimension- and objective-robust, end-to-end Reading-2 witness.

The phrasing matters. This is not a Navier-Stokes existence/smoothness claim, not a Clay-problem-adjacent claim, and not a claim that Sundog has produced original PDE mathematics. Any public copy that elevates this witness into a theorem-facing posture silently retires the discipline that makes the result interesting.

In Plain Words

The body roams; its shadow still decides.

Picture the flow as a body — the full swirling fluid, hundreds of numbers — and a shadow: just its largest-scale features (the "signature"). Sundog's house question, the same one the solar-halo front page opens with and the h-of-x equation makes literal, is whether that shadow carries enough structure to act from when the body is too large to read directly.

Here the answer splits, and the split is the interesting part. You cannot rebuild the full fluid from the shadow — genuinely different bodies cast the same shadow, and we certify it with hundreds of thousands of matched "twin states." But you can make the one safety decision the controller cares about — is a low-band energy spike coming? — from the shadow alone, and get it right almost everywhere. State-insufficient, yet control-sufficient. It is like reading the time off a shadow without being able to rebuild the clock.

New here? That two-part move — read the shadow, then name exactly where it breaks — is the whole site in miniature; the generality umbrella runs it across six harder substrates. Reviewing this? The precise statement, the clause-by-clause claim ledger, and the glossary to standard observer-theory language are in the Proposition.

What The Witness Is

Reading 2 = state-insufficient AND control-sufficient at a matched signature radius.

Reading 2 is the regime where the standard determining-modes signal fails to reconstruct full state, but a smaller Postulate-1 signature nevertheless suffices for the control objective. The witness is constructive: pairs of trajectories that agree on the signature to within ε_K = 0.0664 also agree on the proxy-action objective, at a level that no random-label control reaches.

The G=200 read used the fixed-percentile objective at d = 32, with witness slope 0.736 and a_mm = -0.00079. The G=300 read used the portable quantile objective at d = 18, with a_mm = +0.00058 through the origin and a damp fraction of 0.2688 — well above the v6 0.004 vacuity that motivated the regime-generality pivot. The portable objective is a new objective, so the v5 witness had to be re-established under it before the G=300 test could be admitted; that re-establishment held, with G=200 numbers near-identical to the v5 old-objective read.

The G=300 twin-state lock is certified: TWIN_STATE_CERTIFIED, 100% witness coverage, 942,834 unique pairs, with δ_H = 0.0111 against a real median ‖Q_K‖ = 0.222. The cell-set spec, integrator energy-conservation self-test, and pre-registered verdict-rule completeness all closed cleanly before the verdict was interpreted.

The Mechanism, Measured · Named Novelty: MZ

Why the shadow can decide — a Mori–Zwanzig energy budget that closes on the signature.

A witness that something works is stronger once you can say why. The why here is an energy-budget reading, and it is where we name the explicit novelty.

Split the low band's energy tendency into a part fixed by the signature and a leftover coupling to the unresolved high modes: dE_low/dt = g(Φ_K) + R. A short lemma we can prove says the band cannot feed its own energy — the signature-only transfer T_LLL ≡ 0 — so R is the entire inter-scale transfer, all of it high-mode mediated. The naïve hope, "R is small," is therefore false by conservation.

The measured fact is subtler and stronger: R is ~99% a function of the signature (held-out R²(R | Φ_K) = 0.998 at G=200 and 0.990 at G=300, with calibrated positive and negative controls) — even though the high-mode state is certified not reconstructable from the signature. The flow's state roams; its decision-relevant energy budget is slaved to the shadow.

This is explanatory, not promotion-bearing: it grounds the existing numbers, it does not move C1's status. Detail in the MZ energy-budget diagnostic, framed against standard observer theory in the separation statement (§7).

The Honest Limitation

Marginal, not sharp — and the page says so first.

The separation is real and certified but physically marginal. The signature reconstructs the energy-carrying content of the state almost completely — fraction of variance explained 0.997 in the energy norm and 0.993 in enstrophy — so K = 3 sits near the determining-modes threshold, and the genuinely under-determined part is dynamically-negligible dissipation-range noise. 2D Kolmogorov flow at moderate Grashof has a low-dimensional attractor; the body barely resists its shadow.

We tested the cheap escape — re-reading the gap in the enstrophy norm at G=200 — and it failed (still marginal). A genuinely non-marginal separation, where physically-relevant content is under-determined, needs a higher-dimensional regime (much higher Grashof, 3D, or a different substrate), and that route hits the same numerical wall as the Sabra-shell C2 lane — it needs an adaptive integrator. Stated plainly so a reviewer never has to extract it: this is a weak instance of the regime-2 target, not a gross state/control gap.

Adversarial Robustness · 2026-05-31 · we tried to demote it

Is the shadow only deciding because it nearly rebuilds the body? We pre-registered a probe to find out — and it survived.

The honest limitation above invites the sharpest objection a reviewer can raise: if the signature reconstructs ~99% of the energy-carrying state, then "control-sufficient" might just be "near-determining" wearing a different hat — the shadow decides only because it almost is the body. So we pre-registered a probe designed to demote our own result, with the verdict fixed both ways before it ran.

The test: does control-sufficiency simply track how predictable an objective is from the signature, across a slate of six objectives spanning the resolved band down into the dissipation range? If it did, the separation would be a closed-subspace artifact. It does not. The signature predicts every objective well — fraction-of-variance 0.76 to 1.00, including the high band and the dissipation range — yet control-sufficiency does not track that predictability (rank correlation −0.75 and −1.0, the opposite of the demotion hypothesis). The decision-observability split is not the near-determination in disguise.

And the page flags the one loose thread first. The slate also surfaced an unexpected, still-open puzzle: a dissipation-weighted objective (palinstrophy) the signature predicts perfectly yet is not control-sufficient for — predictable but not controllable, the reverse of the usual story. We do not yet know whether that is a real phenomenon or a burst artifact; it is logged as an open thread, not smoothed over. This is robustness, not promotion: it strengthens the framing for review; it does not move C1's status.

Detail in the objective-overlap discriminator (pre-registration + result §12).

Hypothesis · tested · falsified · re-aimed

We took the split to a resistant body. It wasn't resistant. So we build one that is.

The hypothesis was falsifiable, so we falsified it. The separation measured here is marginal only because the 2D-fluid attractor is low-dimensional — the signature nearly reconstructs the whole state, so there is little "state" for the decision to be unobservable from. The test was to carry the same reconstruction measure (does the low-dimensional shadow rebuild the full state?) to a substrate whose body genuinely resists its shadow. The candidate was Mesa's entangled, decomposition-resistant 5-dimensional net.7 subspace.

It came back marginal too. Ported directly, the measure showed the 256-wide net.7 is effectively ~2-dimensional — it is a function of a six-number observation — so the 5D shadow reconstructs ~99% of it. Mesa is sharp on a different axis (its subspace resists being decomposed into simpler features), but on the state-reconstruction axis this split actually needs, it is marginal, like the fluid. The strongest dynamical-system candidate — a Sabra shell model with a small-scale objective — came back the same. Three for three.

The reason is structural, and it says where to build. A sharp split needs a body that is high-dimensional by construction — not a turbulent cascade that concentrates its content in a few slaved modes, and not a controller whose input is six numbers. That points away from physics substrates and toward a control system whose state is genuinely large. So the next build heads to a language-model substrate — the chat lane — where the body (the model's internal state across a conversation) is high-dimensional by design and the shadow (a maintained ledger) is where we ask whether the same decision-observable-but-state-unobservable split finally turns sharp. This is the measured reason for the pivot, not a result on the model; the high-dim test is still to be scoped and run, and the ∞-dimensional NSE-attractor analogue likewise remains separately hypothesized, cited but not applied.

Falsification Surface

Five named failure modes — what would falsify the coupling claim.

The coupling between Sundog's Postulate-1 discipline and the open Navier-Stokes front is staged on two fronts (a defensible reading leg and a horizon empirical leg) and can fail in five named modes. Each candidate work item declares which modes it attacks.

Next Gate

External PDE reviewer + derived objective selector.

Promotion criteria for the Navier-Stokes ledger require all four of: (a) Front-A vacuity rebutted; (b) runnable Phase-0 deliverable with pre-registered cell set and matched-seed baselines; (c) external-mathematician sanity-check path named; (d) pre-registered failure boundary fixed before numerical reads are interpreted.

The current witness materially advances (a) — a concrete regime-2 witness shows the Postulate-1 reading is non-vacuous on a real substrate — and closes (b) at the procedure-and-execution level. (d) is closed by construction (the cell set, integrator self-test, and verdict-rule completeness are all locked before interpretation). (c) is open and review-gated: an external PDE reviewer pass on Front-A is the next promotion step. The reviewer-facing packet now arrives with the most likely objection — whether the split is merely near-determination — already tested in the open (see Adversarial Robustness, above), so the ask is a focused sanity check, not an open investigation. Final (d) close is coupled to (c). The full twin-state lock receipt and proxy-faithfulness items remain open as part of (c).

Until (c) closes, this page carries the review-gated badge prominently and does not collapse the witness into a positive theorem-facing claim.