Passes 2847–2853: protected observer, noisy M36, and active diagnosis evidence

[wilcompute]

Scope

This PR exists to obtain and retain observable evidence for the Passes 2847–2853 source packet already published on master.

Exact results

• fixed 52-column protected support trajectory punctures optimally to 28 taps at minimum distance four;
• a changed affine-square measurement schedule reaches minimum distance four in 24 samples;
• distance-four observer-word digraph has automorphism order 32;
• the 48 minimum fast selectors have automorphism order 6912 and structure S3 x (S4 wreath S2);
• adaptive no-reset support diagnosis identifies every frame in at most 4 operations, with uniform mean 94/27;
• the phenomenological noisy deep-M36 recurrence has exact saddle node g=(7-3sqrt(5))/4, p=(3-sqrt(5))/2.

Modelled boundary

Maximum-likelihood decoding is tested under explicitly synthetic independent false-positive/false-negative support channels. It agrees with Hamming decoding in the symmetric case and reduces observed word errors by 72.0–76.7% in the asymmetric profiles. No detector calibration or laboratory threshold is inferred.

Evidence workflow

The pull-request workflow:

1. recomputes and verifies the exact certificate;
2. runs focused regressions;
3. simulates directed one-bit correction with Icarus;
4. synthesizes and places the 24-bit encoder and sequential nearest-neighbor decoder using Yosys/nextpnr on HX8K;
5. integrates w33_paper.tex, photonic_holonet.tex, holonet_machine_blueprint.tex, and docs/index.html;
6. compiles all three canonical PDFs with Tectonic;
7. commits generated logs, canonical sources, registry closure, and PDFs back to the PR head.

Do not merge until the dedicated workflow is green and its generated evidence commit has been inspected.