Research consolidation: bind archived inventory evidence and reject optimized false PASS

[azide0x37]

Change and mathematical scope

Research consolidation; no new κ. The archived 146-profile asynchronous inventory bound has a reproducible false-success case: deleting every rank-one charge and running its verifier with Python -O still emits PASS, despite a moment-minus-one margin of approximately −0.30861318239717017. The assertion-enabled verifier correctly rejects this input. A nonexistent source path also passes because the old arithmetic verifier does not read that metadata; this is a provenance gap, not an arithmetic counterexample.

The committed evidence is valid: all 146 profile/width pairs match their retained constructor reports, and both original exact receipts reproduce byte-for-byte. This additive companion pins that evidence chain, independently enumerates the 348 legal patterns per engine, checks both exact duals and the six-step outer recurrence, and retains rejection checks under optimization. All archived files, the selected result, and upstream/ remain unchanged.

The companion uses the two inherited interval engines. It is not a third interval implementation, a new supplier admission, an impossibility theorem beyond the fixed inventory/menu, or a complete multiplication proof. All constructor/physical, all-size, realization, hosting, precision/tape and analytic obligations retain their original scope. The source reports are explicitly PASS_LOCAL_CONSTRUCTOR_ONLY.

Exact base: 66849e1475c930fd7732b1a0067ffee7e1e95f60.
Archive-bound source commit: 01ad6ae33543b4df9732e7399a21f115d5cf94fd.
Integrated PR477 source: a5b588d7f16be383ec80e6c5adaa1a75049cb758.
SOURCE-PINS.json binds the remaining source bytes and archive manifest.

Validation

uv run --offline python -B research/async-inventory-reconciliation/verify.py

Expected: PASS_ASYNC_INVENTORY_RECONCILIATION 146 source rows, 696 dual inequalities, 24 rejection controls, 12 historical replays.

• Twelve distinct controls are rejected for each engine, including omitted charges, infeasible duals, changed profiles/widths/provenance, and altered receipts.
• Seven focused tests pass normally and under -I -O; the dual-pattern count also agrees with an independent generating-function count.
• Reports agree on Python 3.10.21, 3.13.15, and 3.14.7. A committed-files-only export passes from an unrelated directory, including optimized Python 3.10/3.14 replays.
• The original archive-integrity check, selected-record/source-inventory check, and whitespace check pass. No repository-wide supplier rebuild was run for this additive research package. Hosted CI is separate from the local macOS evidence.

OS-random seed 95, research consolidation. No pinned coordination issue existed; CLAIM.md records the narrow scope, candidates, timing, and completed result. The final overlap refresh through #732 found no matching work. #695 concerns selected Linux receipts; #704/#721 concern the central review ledger; #630 corrects the archived E6 screen; #724 audits R/R0 accounting. This contribution depends on none of those pending changes.

Credit remains with Rohan Arun's archive/search/dual construction, Douglas Colkitt's integration and independent base-two arithmetic, PR361's Mercator engine, PR501/502 and PR505, DreamingOfClouds, and the retained predecessor lineage. Prepared by Alexander Templeton (azide0x37) with substantial OpenAI Codex assistance, under Apache-2.0. No independent human peer review is claimed.