M6d: axisymmetric relief, and reconciling the docs with the gates

CONTRIBUTING.md

@@ -58,7 +58,11 @@ When you change or add a model, update the docs in the same PR:
   line at the top of the section. A green test that asserts a known disagreement
   is a `pinned` row, not a `validated` one; if you cannot say which of the four
   statuses your new gate has, the gate is not finished. An asserted constant with
-  no test behind it gets an `ungated` row rather than no row.
+  no test behind it gets an `ungated` row rather than no row. The census and the
+  four statuses are **enforced** by `tests/docs.rs`, which fails with the
+  corrected line — but only the counts are mechanical. A row's *number* is
+  checked by nobody: the ground truth for it is the `assert!` in the gate the row
+  names, so quote that, and re-read it whenever a default changes.
 - **`docs/M*_SPEC.md`** — per-milestone pre-spec gates and design records. New
   milestone work is spec'd here before it is implemented.
 - **`README.md`** — the milestone status table and the CLI examples. A finished

docs/M6A2_SPEC.md

@@ -68,7 +68,7 @@ Stated up front, in the D5 spirit that M6a learned and M6c applied.
 
 **Not gateable: the position of the `P_ig` curve.** Whether a given `Cn²`
 ignites the air depends on M6a's absolute breakdown threshold, which is M6a's
-explicitly ungated quantity (4.8–7.0× above the measured Thiyagarajan & Thompson
+explicitly ungated quantity (3.90–4.69× above the measured Thiyagarajan & Thompson
 curve, inside the 3–10× inter-lab scatter). Every `P_ig(Cn²)` this rung produces
 inherits that offset. The curve's location on the `Cn²` axis is therefore **a
 statement about the model, not about the world**, and must be labelled so

docs/M6C_SPEC.md

@@ -225,7 +225,7 @@ per hydro step dt:
 
 The velocity gates are independent of M6a. The **ignition time and position**
 are not: they come from `AirBreakdown`'s absolute threshold, which is M6a's
-explicitly ungated quantity (4.8–7.0× above the measured T&T curve, inside the
+explicitly ungated quantity (3.90–4.69× above the measured T&T curve, inside the
 3–10× inter-lab scatter). So M6c can pass every velocity gate while lighting
 the spark at the wrong intensity. This is a stated limitation, not a blocker —
 `D` depends on the absorbed `S` at the front, not on where ignition happened —
@@ -450,6 +450,25 @@ velocities, and the spec says so in advance rather than discovering it:
 - Radiation losses and incomplete absorption push the same direction and are
   also out of scope (§ NOT in scope).
 
+**Amended (M6d): the first bullet is no longer available as an excuse.** Radial
+relief is modelled, and its size is measured and pinned:
+`δ = 1 − D/D_wide` = **0.230** at `R_b·α` = 3.2 and 0.305 at 1.6, i.e. a
+finite-diameter beam costs roughly a quarter of the front speed. That is a real
+effect and it is **not the whole ~2× gap** — so relief is part of the answer,
+not the answer. The remaining candidates are the ones already named here
+(radiation losses, incomplete absorption) plus the production EOS and, new with
+M6d, the assumption that the beam travels in straight pencils and is not
+refracted by the plasma it creates.
+
+M6d also found something that changes how any future G7 comparison must be
+made: the modelled front is **transversely unstable**, so a single run's speed
+carries the instability's signature and a comparison against a 1-D calculation
+would attribute that to relief. See [M6D_SPEC.md](M6D_SPEC.md) § G14.
+
+**G7 therefore remains ungated for exactly one reason: there is no anchored
+measured dataset.** That is open question 1 below, inherited by M6d and still
+unpaid.
+
 So the honest claim M6c can make is: **the 1-D model agrees with CJ/Raizer where
 that theory applies, and the gap to experiment is in the predicted direction and
 of the predicted order, for reasons the model has explicitly excluded.** That is
@@ -485,7 +504,11 @@ is labelled as such.
   here, it is unreachable from this solver. It needs a non-paraxial method,
   which is a different solver, not a `Medium`.
 - **Radial / quasi-1-D expansion.** Planar first. This is also precisely why
-  G7 is ungated and expected high.
+  G7 is ungated and expected high. **Retired (M6d).** Axisymmetric `(r, x)`
+  hydro landed in `src/euler2d.rs`, and the relief deficit is measured and
+  pinned at 23 % of the front speed — see [M6D_SPEC.md](M6D_SPEC.md). The bullet
+  stays here, struck through rather than deleted, so the record of what M6c did
+  not do remains readable.
 - **Runtime Mutation++ FFI.** Offline tabulation only (D8/P3). Re-opened only if
   the frozen LTE table provably fails.
 - **Non-LTE / two-temperature plasma**, and finite-rate ionization kinetics.
@@ -495,11 +518,14 @@ is labelled as such.
 - **LSC and LSR regimes.** LSD only; the regime is asserted, not assumed.
 - **Recombination / afterglow / multi-pulse.**
 - **2-D/3-D hydro.** The propagator is 3-D; the hydro is not, by construction.
+  **Narrowed to 3-D by M6d**, which made the hydro axisymmetric. 3-D remains out
+  of scope: axisymmetry assumes no azimuthal structure.
 
 ## Open questions
 
-1. **Which measured LSD dataset anchors G7.** The M6c counterpart of M6a's
-   digitized anchor, and the one input this spec cannot supply itself. Needs
+1. **Which measured LSD dataset anchors G7.** *Inherited by M6d and still
+   open — and now the **only** thing keeping G7 ungated.* The M6c counterpart of
+   M6a's digitized anchor, and the one input this spec cannot supply itself. Needs
    the same treatment
    `tests/data/tt2012_*.csv` got: a named paper, the figure number pinned in the
    provenance header at digitization time, and the setup quoted so the solver's
@@ -547,7 +573,21 @@ is labelled as such.
    what *sustains* it — which is the known experimental situation, where LSD
    waves in clean air are started on a target, on an aerosol, or by a separate
    spike. M6a's ungated absolute level does not touch the conclusion: the gap is
-   10⁵ against a ~7× uncertainty. Pinned by
+   10⁵ against a ~7× uncertainty.
+
+   **Amended (M6a, 2026-07-30/31): the numbers above are the ones this milestone
+   landed with, and the shape of the claim has since changed.** The
+   distribution-resolved closure removed the hard `ε_∞ = U_i` cutoff, so a longer
+   pulse now buys *something* rather than nothing, and seed production raised the
+   short-pulse end. The threshold is therefore **asymptotic rather than flat**:
+   8.815×10¹⁵ at 6 ns falling to 6.745×10¹⁵ by 1 ms, a bounded 1.31× fall that is
+   flat to 1 % over the last two decades. The focus figure is now 6 % over a 500×
+   range, and the level uncertainty is 3.90–4.69× rather than ~7×. **The
+   two-stage argument is untouched**, which is the only thing this step depended
+   on: a bounded fall to a floor is still an intensity criterion, and the drive
+   still sits ~10⁵ below it. What would break it is a threshold falling without
+   limit, and the gate below asserts that it does not. Current numbers live in
+   `docs/MODELS.md` § "The `lsd` demonstration run". Pinned by
    `the_sustaining_drive_is_far_below_the_breakdown_threshold`.
 5. **How hot the run is allowed to get** — *new, and answered (step 4).* The
    `Z̄ ≡ 1` ceiling recorded under "Property closure" bounds where `α_IB` can be