[dillona]

PR #4476 explicitly leaves the fully expanded rank-mass identity from zxv’s #4436 outside its width/routing audit. This package independently checks that boundary, retaining every source complement, paid exit and integer batch multiplicity. It corroborates B_n n − P_n = Δ Q_n and U_n ≤ Q_n ≤ B_n n/Δ for positive surplus; no new construction or κ is claimed.

The separate ordered-word flow and counted expansion reproduce all five published totals for both toy compilations through rank 100 and raw p7 through rank 1,000. Forty-four small programs are also traversed literally, and 14 semantic controls reject omitted complements, wrong multiplicities, missing instances, nonintegral batches and related faults. The universal induction and bank-count qualifications are in PROOF.md.

Validation: Python 3.12.15 registered replay and metadata/pins pass; committed-byte/source-history preflight and checked 19-file export pass; the exported verifier passes with -I -O -B, no Git and PATH=/nonexistent. Changed-source and changed-receipt controls reject. Native/physical execution, record grammar, precision constants, exponent convergence and outer theorem admission remain unverified. The historical orientation-helper failure remains recorded separately; no baseline PASS is claimed.

Sources: #4436 at 56003680aee2871773de408e5f01cc5dd46b144c, #4476 at 4e7022aa9128c8673d3e2024052b2deaa18cb6e4; research baseline 2a13eb1827784206202db90e5dc8327cf8a250db, priority revision 7. Submission was refreshed against canonical main f9f85c1e4fa1e955ff39f10fca447e3e2a4e4adc; the branch includes it while preserving frozen source pins. The research overlap scan covered all 352 open PR bodies and bounded Discussion views, with a fresh submission check for duplicates and newer work. #4483’s changed-height convergence theorem and #4507’s narrower chronological batch widths are distinct from this expanded-instance audit.

Credit zxv for the construction/count theorem, Jacob Alan Sussman for the compiler, Ryanson Jonathan for #4456/#4476 and the explicit coverage gap, Cinna Davis for raw p7, and their retained native/outer predecessors. Prepared by Dillon with substantial OpenAI Codex assistance and separate AI-agent review; no independent human or formal-kernel verification. Only the new scoped package is changed; source exports and notices are retained unchanged.

[dillona]

The hosted planning run stopped in the shared onboarding tests; this package’s replay job was skipped. The four failures are the current-status/CI-navigation mismatches already reported in the #4448 investigation.

I confirmed that tests/test_research_current_status.py, docs/research/current-status.md, docs/ci-verification.md and the selected record have identical Git blobs on this PR’s d72f228237525ad07e8b1bd9000143b464a10225 and canonical main f9f85c1e4fa1e955ff39f10fca447e3e2a4e4adc. The six focused local tests reproduce the same four failures. The registered package replay and its committed/exported checks passed locally as documented. A separate upstream navigation repair and CI rerun remain necessary; this is not a hosted replay pass.