Passes 3175–3186: curvature-routed inference and three-edit epoch

[wilcompute]

Stack

This PR is intentionally stacked on PR #243's readable head e35b90b2535f45e570b782c6c9782ab74eb00d93. It executes the five requested continuations plus two independent outside-box fronts while preserving the Photonic Holonet and machine-blueprint boundaries between selector clock, route bus, address sheets, temporal extension, support/readout, phase/execution and fail-closed M36 injection.

3175–3176 — curvature-conditioned sensing

The 48,826 hypotheses split into 45,445 non-shared, 1,725 shared-flat and 1,656 shared-curved states. The exact chain rule I(H;Y)=I(K;Y)+I(H;Y|K) proves that resolving the D4 curvature latent cannot reduce information.

For the explicit synthetic channel and seed:

• stress regimes: 1/32 actions change; mean best-action gain 0.0355238394 bits, maximum 0.0902425706;
• frozen sparse operational prior: 0/32 actions change; mean gain 3.9383e-6 bits.

Curvature is therefore retained as a diagnostic/postselection coordinate, not advertised as an unconditional operational win.

3177 — all-194 information frontier

All 80 universal five-opcode and 114 universal six-opcode designs are recomputed. Each receives average, minimum, maximum and normalized dispatch information, entropy variance, collision probability and decoder-operation cost.

The exact six-objective Pareto frontier contains eight designs. Global extrema:

• maximum average: 2.4141403814 bits, four attainers;
• maximum worst-frame capacity: 1.7924812504 bits, 22 attainers;
• maximum normalized capacity: 0.9481541058, four attainers;
• minimum variance: 0.0357531815, two attainers;
• minimum collision probability: 1/9, four attainers.

No single design owns every information criterion; this frontier is separate from the full-group BFS runtime frontier.

3178 — optimal three-edit phase epoch

Twelve payload-unused symbols define M_p = u_p^7. Pairwise marker distance and marker-to-payload distance are at least seven, which is optimal for three adversarial edits because d_min >= 2t+1 = 7.

For a received word of length m containing c copies of u_p, d_L(u_p^7,y)=max(7,m)-min(7,c). The exact radius-three ball contains:

• 3,667,012 traces per phase;
• 44,004,144 phase-labelled traces across twelve phases;
• zero required clean payload symbols afterward.

3179 — proof-carrying M36 envelopes

Candidates are canonical content-addressed objects carrying engine/shard/source provenance, independent certification and hashed projector/Pauli/logical/error witnesses. The Z0/Z1/Z2 negative control is validly sealed and rejected. A one-field mutation fails digest verification. No accepted candidate is asserted.

3180 — routed joint utility

The 23 detector triangles form a connected route graph of diameter two and degrees 14–18. The explicit programmable utility combines detector MI, curvature MI, shortest-route distance and multiplicity, epoch confidence, ISA capacity and runtime cost.

Across 64 deterministic scenarios:

• detector-only action changes: 8;
• mode choices: fast6=45, low4=11, current4=8;
• mean modeled utility gain: 0.0345842444;
• maximum modeled gain: 0.0892247164.

These are exact outputs of the explicit synthetic model, not physical calibration.

3181 BONKERS — D4 triangle Wilson flux

Phi(a,b,c)=kappa(a,b) xor kappa(b,c) xor kappa(c,a) is invariant under simultaneous D4 conjugation. Exact census over 343 ordered nonidentity triples:

• flat flux: 223;
• curved flux: 120;
• simultaneous-conjugation orbits: 106;
• across 23 measured triangles: 5,129 flat and 2,760 curved local assignments.

This is an algebraic holonomy syndrome, not spacetime curvature or measured optical phase.

3182 BONKERS — recursive belief virtualization

A typed live context is 52 bits. At level n, global replication costs 52(40^n-1)/39 bits while one active root-to-leaf path costs 52n bits with routing diameter at most 8n.

At level six:

• globally replicated state: 5,461,333,332 bits;
• active-path state: 312 bits;
• exact virtualization ratio: 17,504,273.5.

This assumes one active path; concurrency and checkpoint traffic remain separate.

Source and evidence

The PR contains seven exact/model engines, a focused closure, typed report and claim ledger, source summary, canonical LaTeX/site inserts, idempotent integrator, unified RTL contracts, protocol testbench, pytest regressions and an Icarus/Yosys/HX8K/three-PDF workflow.

Do not merge before PR #242 and PR #243 are green and merged, this PR's focused lane is observed green, and the generated outputs are inspected. No claim is made for observed FPGA area/timing/power, compiled PDFs, exhaustive M36 outcome, laboratory likelihoods, edit rates, heat, coherence or optical 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]

Execution and evidence boundary:

• All seven exact/model fronts have been independently regenerated by GitHub Actions. Focused pytest passed 6/6.
• The first external run also passed the unified Icarus protocol test and synthesized/placed all four representative HX8K tops: curvature accumulator, three-edit epoch decoder, D4 Wilson-flux unit and streamed routed-utility argmax.
• That run failed only when the idempotent integrator decoded legacy manuscript bytes as strict UTF-8. Commit 932c7e9d8c401451962fb93730a9e2289766cfab replaces strict decoding with lossless surrogateescape round-tripping; no source theorem or numerical result changed.
• Rerun 30920672660, job 92030123800, is in progress from the corrected head. Paper integration, three PDF builds and artifact upload remain unobserved until that run finishes.
• PR Passes 3153–3162: larger ISA census, epoch delimiter, and factor engine #242 is separately rerunning after its Git-identity workflow repair. PR Passes 3163–3174: proof-carrying Bayesian runtime and non-Abelian frontiers #243's readable 32-file source tree is present, but its newly installed exhaustive workflows still require repository approval.
• Do not merge or claim compiled PDFs, observed final paper integration, exhaustive M36 outcome, global full-BFS physical optimum, FPGA power, laboratory likelihoods, physical edit rates, heat, coherence or optical performance from these source/digital results.