Passes 3214–3225: all-194 runtime, quotient ROM, proof roots, and finite-port cohomology

[wilcompute]

Scope

Executes the five outstanding machine fronts plus two independent outside-box constructions against current master, including merged Passes 3205–3213. The packet is independent of the still-open #242 -> #243 -> #244 stack and does not promote its queued or action-required evidence.

Five requested fronts

3214 — complete all-194 affine runtime surface

A canonical plan recomputes all 80 universal five-opcode and 114 universal six-opcode designs, including collision, frame-distance, operation-cost and destination-information metrics. Thirty-two exact shards perform directed BFS on all 4,199,040 elements of ASp(4,3).

The aggregate refuses missing or duplicate shards, cross-plan records, stale plan hashes, duplicate global indices, incomplete growth series, or any design whose reached order is not exactly 4,199,040. Only a complete 194-record aggregate emits the joint runtime/information Pareto frontier. No global winner is claimed at source publication.

3215 — observed tri-ISA promotion contract

Source metrics for current4, low4 and fast6 remain exact for the frozen generators, but the physical promotion path now requires one observed device/toolchain/commit record containing placement, logic-cell and Fmax evidence for all compared modes. Missing placement or calibration preserves current4 as the fail-closed fallback in both Python and RTL.

3216 — canonical 876-state quotient ROM

The compiler independently reconstructs all 48,826 D4 hypotheses, 46,284 base signatures and 1,436 collision classes, then forms complete recursive output/action/child signatures. It assigns deterministic IDs to 770 initial and 876 total Moore states and emits a 102-bit ROM word containing eight child IDs, an outcome mask, curvature histogram, action and terminal bit. Equality of complete recursive signatures is the bisimulation certificate.

3217/3222 — proof-carrying runtime and M36 evidence

Runtime and M36 shards become domain-separated leaves binding plan, engine, source, result and candidate identities. Separate Merkle roots, an umbrella root, deterministic audit challenges, inclusion proofs, tamper controls, duplicate-index rejection and actionable naysayer violations are included.

This is integrity/provenance composition, not a proof of honest computation, zero knowledge or consensus. Full recomputation and independent certification remain authoritative.

The M36 adapter materializes rather than reimplements the Pass-3125 engine. All 256 unique shards are necessary but not sufficient: examined coverage must sum exactly to 50,868,675 before zero candidates can become a no-go under the frozen engine.

3218 — exact stack-drain preconditions

The frozen live snapshot records:

• Passes 3153–3162: larger ISA census, epoch delimiter, and factor engine #242 open, non-mergeable against advanced master, focused rerun queued;
• Passes 3163–3174: proof-carrying Bayesian runtime and non-Abelian frontiers #243 stacked on Passes 3153–3162: larger ISA census, epoch delimiter, and factor engine #242, workflow action required;
• Passes 3175–3186: curvature-routed inference and three-edit epoch #244 stacked on Passes 3163–3174: proof-carrying Bayesian runtime and non-Abelian frontiers #243, focused evidence unobserved;
• Passes 3193–3204: month-wide claim audit, typed runtime fusion, and seven-front closure #246 independently open, focused lane pending;
• Passes 3205–3213: exact port geometry and chromatic proof surface #247 merged source with PDF/physical boundaries retained.

The scheduler validates the DAG, emits the exact dependency sequence, reports zero merge-ready PRs and grants no merge authority.

Outside-box front A — finite-port sheaf/cohomology lift

The merged 45-block/240-port factorisation defines a cellular complex with:

(V,E,F) = (45,720,240).

The 240 triangular face boundaries have disjoint edge supports. Over F2, F3 and F5:

rank d1 = 44, rank d2 = 240, (b0,b1,b2) = (1,436,0).

A phase edge assignment is flat only when d2^T a = 0. Nonzero port flux obstructs a flat lift, while zero flux leaves p^436 gauge-inequivalent constant-coefficient lifts over Fp. The failed direct 45-object bridge therefore becomes a large moduli problem, not a phase-bit repair. This is not claimed as a complete contextuality invariant.

The induced binary CSS complex is exactly:

[[720,436,2]], with dX=2, dZ=3.

The block graph has 5,280 triangles; only 240 are filled port faces, leaving 5,040 weight-three non-boundary cycles. Distance two makes this a high-rate provenance/checksum fabric, not a fault-tolerant memory.

Outside-box front B — synchronizing rank and reset thermodynamics

The curvature controller crossed with twelve phase states has 876 x 12 = 10,512 states. Every phase-only word acts identically on belief, so its image rank is at least 876. The optimal marker attains rank 876 by collapsing phase, proving that epoch synchronization is not epistemic reset.

A separately authorized rank-one reset removes the full finite-state capacity log2(876) = 9.7747870596 bits. The Shannon entropy actually erased depends on the input distribution. For a uniform or maximally mixed 876-state ensemble at 300 K, the Landauer floor is 2.8063207254e-20 J = 0.1751567627 eV. This is not a distribution-independent per-event charge or an engineering energy estimate.

The RTL lets a marker lock phase but requires separate reset authorization and a matching proof root before emitting a belief-reset pulse.

Execution lanes

This PR adds three guarded workflows:

1. Focused lane: regenerate fast exact artifacts, pytest, adversarial Icarus simulation, Yosys/HX8K placement of the quotient ROM and control gate, idempotent integration, and all three canonical PDFs.
2. All-194 lane: 32 exact full-affine BFS jobs followed by one strict aggregate and evidence commit.
3. M36 lane: 32 workflow buckets covering 256 logical shards, exact coverage validation, aggregate, proof-root materialization and evidence commit.

Evidence boundaries

Do not merge before the focused lane is observed green and its artifacts are inspected. The exhaustive workflows may complete independently, but no aggregate may be described as complete before its own terminal artifact exists.

No claim is made yet for:

• a complete 194-record runtime aggregate or global larger-ISA winner;
• a complete M36 census, zero-candidate no-go, or accepted candidate;
• observed tri-ISA area/timing/calibration;
• observed quotient-ROM FPGA area or Fmax;
• successful three-PDF compilation;
• laboratory optical loss, error rates, power, heat, detector, coherence or fabrication performance.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[wilcompute]

Evidence boundary at publication: source and exact finite constructions are committed. Focused run 30935184646, all-194 run 30935186904, and full-M36 run 30935184026 are queued in the repository-wide Actions backlog. None is treated as green. The PR must remain unmerged until the focused lane is terminal-successful and its generated ROM, RTL, placement, integration, and three-PDF artifacts are inspected. The exhaustive aggregates retain their own independent completion gates.