[dillona]

This adds a source-bound causal-history split with conditional κ = 0.003642959019031, a gain of 0.000000550522162 over #7131. Seven timed gathers, two explicitly owned buffers and one later injection split a rank-nine helper into ranks four and five. One paid path changes from rank six to ranks one and five; all added storage and scalar work is charged. Ali Romman supplied the causal-history mechanism; this package supplies the new exact instance and complete application.

The complete word adds 27 ADDs, four SCALEs, one MOVE, two INITs and two CLEARs. Owned inventory changes from 51 labels/54 lifetimes/peak 43 to 53/56/44. The 17-stage application has 110,417 paid positions, 22,325 productive fields, surplus 1,424, zero empty credit, ordinary source-unit work 779,064,884 < 10⁹, and precision c=2362 on the freshly derived 1125/F25 grid. The original caller is byte-identical.

All three direct CPython 3.14.7 commands and the final registered isolated CPython 3.12.16 replay passed, matching every frozen receipt field. All 23 scientific hashes stayed unchanged. Coverage includes the source-reconstructed timed coefficient certificate; full forward/adjoint and COPY/lifecycle checks; 18 native corruption controls; phase and five 512-column chart checks; finite physical replay; all 3,039 paid/helper maps and 104 main boundaries; both directed prices; fresh owned-stock, ordinary, precision and compiler/fixed-tape costs; and 54 strict + 4 functional outer conditions. The first native attempt stopped on the old fixed lifecycle census. Counted adapters now retain its exact check against (56,44,53), and an extra uncharged lifetime is rejected. Frozen source bytes are unchanged.

python3.12 -B scripts/research_package.py --package research/p104-causal-buffer-split

The complete 37-file independent composition artifact at 666fd749ec465de7238633a0c2fd5ee3081477e0 and all nested sources/notices are frozen. It contains the same canonical source word as Rohan Arun's completed #7131. Candidate native SHA256: 81f9b2e0a170aa3ab0cd4bb1056c834ebe0b98eaf37abe5986c6e53abd1257ea. Discovery stopped at its first positive, after 150 of 7,687 individually affordable paths; no global temporal ceiling is claimed.

Credit Ali Romman, Jerry Chen/CubeJerry, Matarener, Rohan Arun (#7122/#7124/#7113), Cinna Davis/1337isnot1337, Dugongue, ToikY, DreamingOfClouds, Douglas Colkitt/#7000 reviewers, the dillona #7130/666fd predecessor artifacts, and all retained source authors. Prepared for dillona with substantial OpenAI Codex assistance; no subagents. Apache-2.0.

The five inherited framework assumptions, independent human/formal acceptance, maintainer admission and compiled top-level execution remain open. This is a conditional finite construction; the selected frontier is unchanged. Hosted CI is separate from the completed local scientific replay.

Canonical investigation and overlap record.

Task p104-side-correction, with the premise changed from a simultaneous cut to timed collection/delivery and explicitly charged owned buffers. Canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8 and selected #7000 were refreshed unchanged before submission. The complete 23:47–23:48 UTC overlap snapshot and the publication-time new-PR check found no competing current-source construction or stronger complete κ claim in that coverage; sparse local coverage is explicitly limited.

Open next questions: independently replay the timed rows/column and the two owned-lifetime cost substitution; then consider a distinct fully priced temporal domain on the new source after refreshing guidance and overlap.

Hosted status at public head a95783a: run42928053888 stopped in onboarding: 760 tests, one failure, 24 skipped. CurrentCiNavigationTests.test_current_ci_section_matches_selected_record expects make frontier7000-verify, while the unchanged CI guide still points to #6030. The scientific replay job was skipped. This is separate from the completed direct and registered local scientific replays; no hosted scientific pass is claimed.