[rohanarun]

This raises conditional κ to 0.003643216822476, a gain of 0.000000257803445 over #7140 (a95783a8c861248a759136d4285ea725499cf231). A charged input-boundary extension makes one more exact helper split possible on the causal-buffer source.

Helper 1479’s input moves to frame 3, with its extra rank-one transport charged. Retiming one contact-free path then splits a rank-nine helper into seven and two. Combined native deltas are T2 +1, T7 +1, T8 −1; P1 +1. The emitted word adds 666 ADDs, 4 SCALEs and 1 MOVE; owned inventory remains 53 labels, 56 lifetimes and peak 44. All 17 native uses are included in the rebuilt caller bill.

The complete application has 110,434 paid positions, 22,342 productive fields, surplus 1,424 and zero empty credit. Ordinary source-unit work is 779,695,108 <10⁹. Fresh precision increases from 2362 to c=2380, with independently derived native grid 1125/F25. The caller is byte-identical to its retained source.

Direct source-aware checks and the registered CPython 3.12 replay pass. Coverage includes whole source/candidate reconstruction, arbitrary-input forward/adjoint, ownership, six 512-column native chart/phase checks, ten native corruption controls, finite physical replay, all 3,039 paid/helper maps and 104 main boundaries, both directed prices, fresh grid/envelopes/ordinary/fixed-tape costs, and 54 strict + 4 functional outer conditions. Every frozen receipt field matches; all 20 scientific file hashes are unchanged across registered replay. See REPLAY.txt and VALIDATION.md.

python3.12 -B scripts/research_package.py --package research/p104-endpoint-funded-split

The draft was published immediately after its narrower exact native/price gate passed. Full verification then completed before readiness. The full 30-file #7140 export and all nested sources/notices are retained; no predecessor bytes or selected records were changed. Candidate canonical native SHA256: 3bb4d3a9d8a222af1155c366e52f0b787029b937f71d870f1e407feb1c1d9f33.

Credit Ali Romman’s helper identity, CubeJerry’s contact-free retiming and verification engines, Matarener’s kernel methods, dillona’s source constructions and smooth-pivot basis, ToikY’s pricing, Rohan Arun’s signed caller/recharts, and all retained authors. New paid input extension/composition by Rohan Arun with substantial OpenAI Codex assistance; Apache-2.0.

Five inherited assumptions, independent human/formal acceptance, maintainer admission and compiled top-level execution remain open. Local registered replay is separate from hosted scientific CI and overall CI. The existing PR #7131 CI-navigation correction is retained and its targeted test passes; it does not change the selected result or repair unrelated historical replay drift.

Open next questions: independently audit the paid boundary substitution and exact cut identity, or discharge a named inherited assumption. This finite search makes no global optimality claim.

Hosted scientific replay passed on exact head 316b93f25361c0fd1b3a50e1180bd59096c3e45e: scientific job 130235385452. The hosted log records the early exact gate, full native check, complete application, and PASS_REGISTERED_RESEARCH_PACKAGE at 00:46:57 UTC on November 11. The overall workflow is red because its separate historical-selected-replay-gap job reports working source drift: Makefile; that is distinct from this package’s successful scientific replay. Five inherited assumptions, independent human/formal acceptance, and maintainer admission remain open.