Remove P104 private descents through late FULL dirty-query delivery

[Clear1nsight]

Result

This coupled P104 consumer/cleanup rewrite removes all eight private rank-seven descents while preserving the exact PR3882 endpoint encoding and all 72 arbitrary dirty helper returns. It retains the actual Q producers and every public native transport. At the existing FULL cut, it reuses the correction's 104 target lifts, replaces 32 private deliveries plus 32 paired D cleanup reads with 416 direct FULL reads, restores the quartered query donors, and performs the existing Q source cleanup.

The completed local native mass falls 2576 → 2520. The complete finite scalar difference is +352 unit ADDs and +16 each half/double, with no private allocation, copy, or erase. The raw all-FULL Gram module has mass/progress 2416; this is a changed boundary, not permission to substitute that progress in the old caller gate.

The saving also applies compositionally to PR3894's source-preserving correction through the same all-FULL interface. It is not a gain exclusive to source consumption or role exchange. That compositional statement is scoped separately from the actual frozen PR3882 tail replay.

Evidence and boundary

• Exact source-bound Fraction replay on all 280 independent inputs; every row and all 72 dirty returns match the old completed channel. The actual frozen PR3882 tail preserves all endpoint rows, residual encoding, sum slots, and frame dimensions.
• All 104 retimed lifts have concrete rank-one native recipes. Their ordered canonical entrance bases come from the actual retained terminal Y certificate, not merely equal spans. An independent interpreter checks signed Pauli actions plus global phase for every lift, 512 representative Gaussian columns, and a phase fault.
• Four semantic failures reject: absent late delivery, delivery after query cleanup, an omitted target, and an unrestored donor. Seven actual optimized CLI fixture corruptions reject after each fixture first passes normally. Boolean/float receipt changes, duplicate JSON, false savings, changed/missing source, and a changed frozen tail are covered.
• Registered checks pass in normal and optimized mode. Independent AI peer inspection/replay passed; this is not human or formal review. Full historical native paths and the final correction/global/all-size admission remain inherited to the expressly stated extent.

Under exactly PR3954's favorable corrected two-axis assumptions, its complete gate still fails by 32448: 208*(2520-2312)-10816. The bill is 2997696 against denominator 2965248. No selected frontier, global supplier, all-size result, universal COPY floor, or improved kappa is claimed. The failed gate stops this construction round.

Provenance and publication

Canonical main 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority 6; 11 files only under research/p104-dirty-catalysis/. Fork head cdb7dcf3f7e5f860714a1b12df4c7a5157d8ae3d, tree 8112748fd1ea7b1b8ed3052b8ec1fa927f02ddb1. Exact committed bytes and source-history preflight pass. All five registered commands pass from the official source-complete read-only Git-free export; all 17 exported files retain bytes/modes. Archive SHA256 b439add1dc998d72a3e348d443412a0f63b4dc92af2aba0259ae87a60ed17676. Hosted CI passes at this exact head; both planning and the focused five-command replay succeed, including committed-input preservation. Reproduce with python3.12 -B scripts/research_package.py --package research/p104-dirty-catalysis.

Overlap review used the complete 676-open-PR body snapshot at 07:38:11 UTC, repository helper, and live coordinator board review; main was refreshed unchanged before publication. The helper had no Discussion cache, so the live claim check is separate: #3954 (comment) . This is bounded overlap evidence, not a novelty certificate.

Credit Rohan Arun's PR1972 word; Chafik Boukhalfa's PR1991 native certificate; Adam Gall's PR1940 delivery work; icekylinx PR531 compiler and PR1862 interpreter lineage; Josie Elliston, Sagar Thakrar, Douglas Colkitt, and Dominik Scholz's P104/E8 work. Chris R / Clear1nsight's PR3882 supplies the frozen correction, PR3943 the related earlier-cut cleanup precedent, PR3954 the caller gate, and PR3965 the fixed-consumer obstruction and adapted sparse evaluator. PR3961/3962/3966 provide distinct boundary, coverage, and gauge constraints. The new chronological premise changes the actual consumers; it does not contradict the fixed-consumer predecessor or claim PR3943's separate scalar savings.

Prepared by Chris R / Clear1nsight with substantial OpenAI Codex assistance. No merge, closure, selected-frontier, or shared workflow change.