Bind mixed port coverage to actual endpoint dimensions

[ywz65]

PR2297 supplies 22 orthonormal five-frames and a residual pair, while its favorable screen assumes five unpaid coordinates at every port. This audit preserves the producer's exact binary factorization and explicitly conditional uniform-five calculation, but shows that the literal mixed cover does not supply that premise.

The exact incidence recount gives 110 ports of dimension five and two of dimension two, hence weighted unpaid dimension 277/56. With the retained selected normalization and local profile, the conditional mixed-port screen adds 16/3 paid mass per call and reduces the available rank-one overhead from 157 to 152. The target a=1/99 still passes. A diagnostic native comparison proves the residual pair has four nonzero coefficients per address column, whereas any orthonormal five-frame completion has 32 and changes the terminal translation. Such a completion requires its own construction and invoice.

This is a missing-premise audit and corrected conditional planning budget, not a disproof of the producer's conditional calculation, a new physical word, supplier, kappa, or impossibility of completing the five-dimensional contract.

The retained charge 118 and local profile are counterfactual inputs to this planning calculation, not a certified scalar or operator-valued port decoder. This audit does not promote the binary factorization to either endpoint contract. Decoder-realization failure can invalidate that hypothetical invoice without changing the port-incidence recount; any replacement decoder must be repriced. PR2532 addresses that complementary decoder premise; its new certificate has not been independently replayed here.

Evidence and reproduction

Only research/mixed-port-coverage-audit/ is added. The package binds PR2297 head 49f9c1a08bf1fe67a62bd6d2825aa0d03e1b1518, preserves six predecessor files byte for byte, authenticates source SHA-256 and Git blobs before execution, and runs the original verifier unmodified in a Git-free export. The selected archive binds all 112 labels and the actual four-pair, nine-stage, 16-replica normalization.

python3.12 scripts/research_package.py --package research/mixed-port-coverage-audit --check-only
python3.12 scripts/research_package.py --package research/mixed-port-coverage-audit

Both passed on CPython 3.12.7. Registered checks include exact rational before/after screens, the sharp 152/153 overhead threshold, all 512 native basis-column witnesses, six semantic controls, a relocated optimized Git-free replay, and rejection of actual producer-source, price-source and receipt corruption. Default verification preserves source and receipt bytes. Separate Codex-agent proof/source review found no blocking issue, and its independent CPython 3.12.14 isolated optimized primary replay also passed.

Scope and overlap

Started from and refreshed canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4; no main or producer-head change was found. Picker seed 6bf29e3512c101690f3c7b44cf3ce313 selected recursive interfaces. The coverage-law card changes scope to an audit of this existing mixed-dimensional premise; the prescribed ceiling positive control passed. TASK.json retains the exact original card and explains the change.

The complete latest REST listing covered 720 open PRs at 2026-10-15 13:31 UTC. The 1,197-record Discussion snapshot is from 12:29 UTC and may be stale because the account GraphQL quota prevented refresh. Bounded live Curiosity Bot reads of the relevant threads do not establish complete board coverage. OVERLAP.json records those limits. PR2400 concerns PR2308's different shared pair of frames and its weighted dimension 31/28; weighted averaging is not claimed as a new principle here. This result concerns PR2297's residual two-plane, its endpoint, and its corrected 152-unit allowance.

Credit and next work

Credit Cinna Davis / 1337isnot1337 for PR2297 and its disclosed Codex assistance; the existing plane-port and coverage-law authors, the retained ceiling engine, and the selected construction's source lineage remain credited in the package. Prepared by ywz65 with substantial OpenAI Codex assistance in research, proof, implementation, checking and writing. Separate AI review is not human review. Apache-2.0 attribution and predecessor bytes are retained.

The README gives two bounded next steps: supply and fully price a uniform-five completion of the residual ports, or realize the literal heterogeneous word through the nine-stage native controller. Neither physical realization is supplied by this audit.