Prove a marked-residual stock floor and bind its reflection escape

[RossRichardson]

Summary

• Prove a full-stock native rank-mass floor in the declared whole-plane native, whole-plane ADDRESS, X-translation and Gaussian scalar-bank grammar, including arbitrary dirty helpers restored to identity.
• Exhibit an actual unpaired reflection outside that grammar, with phase-correct native recipes and paid adapters. Cheap source conjugation loses the same retained progress, so this does not give a favorable complete caller.
• Correct the joint-debt query-count corollaries to single-plane, rank-2f queries. Two multi-plane calls can implement diag(N0N1,N0N1), while still paying four planes and rank mass 8f. Only research/marked-residual-reflection/ changes; original proof/checker bytes and historical receipts are retained.

Scope and limits

The typed half-chunk extension has a separate f≥2/query-alphabet boundary. General bit-linear ADDRESS, arbitrary Pauli Z, noncommuting adapters and changed targets remain outside the stock floor. Written all-parameter proofs are separate from finite Gaussian, character, normalizer and support audits.

The general floor measures native-plane count/rank mass, not the number of arbitrary-support unary queries. The P2 review finding is corrected in the proof and assertion message. The registered regression checks the marked-basis actions and phases of two-, three- and four-call full-pair words plus the two-call q3 split. The main rank-mass theorem is unchanged. No favorable complete caller, new κ, physical supplier admission or formal-kernel theorem is claimed.

See the corrected proof, coverage and limits, and task, scope change and open next questions.

Test plan

All new local evidence below is bound to 7b0c74ef85215d438c85b610360c67b55c5f2a15.

• Fresh registered normal/optimized replay: python3.12 -B scripts/research_package.py --package research/marked-residual-reflection. All existing replay stages and the new query-count controls pass.
• Committed bytes, metadata/pins and source history: python3.12 -B scripts/check_research_commit.py --package research/marked-residual-reflection --check-source-history and the registered runner's --check-only mode pass.
• Official package export/integrity and read-only Git-free python3.12 -I -B verify.py / python3.12 -I -O -B verify.py pass with empty PATH; exported bytes remain unchanged.
• Separate AI delta review confirms the single-plane restriction, two-/three-call counterexamples, exact original-file preservation and unchanged historical receipts. AI review is not independent human or formal-kernel acceptance.
• Automatic CI run 39257830364 completed successfully at this exact head, including planning and registered package replay. Full workflow status/conclusion and source SHA were read back; no manual rerun was requested.

Sources and credit

Canonical base 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority revision 6. SOURCES, BINDINGS and RELATED-WORK retain exact ancestry and nearest overlaps; BINDINGS records the correction and both preserved originals. Canonical publication and review/correction update. Live overlap coverage is partial, not a novelty certificate.

Ross Richardson with substantial OpenAI Codex assistance; Apache-2.0. Preserve icekylinx/PR531 and Rohan Arun/kernel92 compiler notices and recorded ancestry; Chris R/Clear1nsight #3987 and ToikY #4002 signed-lift/phase credit; and Clear1nsight's distinct #3989/#4045 interface/volume work. Earlier #3630/#3632/#3645/#3649/#3655 and separate BIT #4021/#4023 remain excluded from these new claims. No merge requested.

[RossRichardson]

Reviewed exact head 3a82d1e1e3b55e6bfcd60e73c8d552ca85bf0fd6 against canonical main 20ecf2891d80b5acf18d980b2164fb44ea7faa90.

One P2 correction remains: the joint-debt corollary confuses total native-plane count with unary bank-query count. The inline finding gives an explicit two-query counterexample and the required single-plane restriction. The main rank-mass/full-stock theorem is not refuted.

Within the stated grammar, separate AI proof review found no additional defect in the whole-plane commutator bound, restored arbitrary dirty-helper extension, typed f>=2 half-chunk boundary, phase-correct unpaired reflection recipes or source-progress identity. The fixed-cut tail remains conditional on its stated factorization. No physical supplier, new kappa or selected-frontier conclusion follows.

Fresh committed-byte/source-history and registered metadata/pin checks pass. Existing CI run 39248164804 was read back completed/success at this exact head. The proof review included a small independent commutator-rank enumeration; the package suite was not rerun. Passing CI does not resolve the prose overclaim.

Size: 48,864 added lines across 57 files; 44,384 JSON lines and 3,722 Python lines, 1,044,492 bytes total. Retain frozen evidence/attribution. The PR description should use the required four-section format, link the canonical finding, and retain the query-count correction as an unchecked gate.

Prepared by Ross Richardson with substantial OpenAI Codex assistance. This is an author-side AI review, not independent human or formal-kernel acceptance. Leave substantive acceptance pending the scoped correction.