[aliromman]

Constructive conditional ordinary BIT saving: 0.003488579527, an exact gain of 0.000000657061 over PR #5648. This composes PR5641 retirements, PR5648 snapshots, and the released PR5647 helper444 return. It resolves their shared-centre conflict and uses the direct SOURCE77 version instead of five historical gathers. No whole-kappa or selected-frontier update is claimed.

Scope, proof, exact checks, source pins, registered replay. Base: canonical main df0137efd35e11b4445734bf612bdd4c7ebc330c, priority revision 7. Contribution head: b13c9ae8a539c05f0f590ce9d08c7067d99c6145.

All 11,166 retained originals and their target debits remain. The complete caller pays 56,799 queries and allocates 17 owned banks with no output credit. Relative to PR5648, delta P=-18 H1-18 H3+18 H4, delta T=0, so the gain is positive for every 0<t<1. Discussion #5652 records the overlap-checked composition claim.

Normal and explicit optimized registered replay pass. The local word tracks all 662 original/dirty columns. Separate adjoint, stream and interval engines check the full 445,876-event caller, query descriptors, target owners, positive fees and four acyclic ordinary levels. Two census implementations check all 667 eligible blocks. A full lifetime check finds only centre658 available for all three overlapping snapshots, so this direct-copy schedule needs two new banks. A separate historical-span audit checks all 2,878 remaining nonzero local queries and finds no direct historical replacement. Raw crossing is in (0.00348857954616,0.00348857954618); the admitted saving is within 1.92e-11. Source, stale snapshot, missing teardown, stale epoch and unfunded-price controls reject.

python3.12 -B scripts/research_package.py --package research/g20-complete-fusion

Python 3.12 and the standard library suffice. Committed-byte checks pass. The native/all-width, ROW, compiler, precision, tape and analytic premises remain conditional. Full historical suites and formal-kernel proofs were not rerun; separate implementations are not independent scientific acceptance. The finite uniform program and astronomical decimal threshold are algorithm-defined, not materialized. Other producers, protected-read retiming and paid frame changes are outside the frozen census. This shallow checkout cannot run the full ancestry audit. A reproduced upstream Current CI missing-link regression currently blocks the generic hosted planner; the package has complete local replay evidence. Hosted plan job121372851509 ran449 shared tests and failed the two Current CI navigation checks for the selected review and source-manifest links. Its scientific replay job was skipped. This is the reproduced upstream main regression; the normal, optimized and Git-free package replays above pass locally.

Credit Dugongue PR5629/5639 for the retirement and snapshot proposals; ai-nanachi PR5512 for native G20 and ordinary admission; Ali Romman PR5421/5641/5648/5647, including the direct SOURCE77 handoff during PR5648. PR5647 at 41c3678 explicitly released composition. Retain DreamingOfClouds, eumemic, Cinna Davis, rohanarun, zxv, Jacob Alan Sussman and all frozen notices.

The three released fusions are now composed and priced. Further gains need a changed producer, protected-read retiming, paid frame map or new storage encoding. The frozen block and direct historical-query domains are exhausted. A compatible stronger complex caller remains necessary for a whole-kappa gain; no further search is reserved.

Prepared by Ali Romman with substantial OpenAI Codex assistance; Apache-2.0.