[dillona]

This composes Rohan Arun's completed #7142 endpoint-funded split with #7143's temporal source, raising conditional κ to 0.003643438506419: +0.000000221683943 over #7142 and +0.000000257834774 over #7143. Both parents and their nested sources/notices are retained. This supplies the exact translated witness and full combined application; it does not claim invention of either mechanism.

The endpoint cut moves to 44,638, and output coefficient −349/32 moves from retired role 648 to role 1449. Every prefix/suffix coefficient is checked again. A charged rank-one input extension for helper 1479 enables the rank-nine split into seven and two at chart 529; role 228's contact-free retiming keeps its paid rank two. Native delta relative to #7143 is P1 +1; T2 +1, T7 +1, T8 −1. The word adds 666 ADDs, four SCALEs and one MOVE. Owned inventory remains 55 labels, 58 lifetimes, peak 44; every existing buffer and added scalar/service cost is charged.

The complete unchanged 17-stage caller has 110,451 paid positions, 22,359 productive fields, surplus 1,424, zero empty credit, ordinary source-unit work 779,890,268 < 10⁹, and fresh precision c=2380 on grid 1125/F25. Both directed prices exceed both parent claims.

All three direct CPython 3.14.7 commands and registered isolated CPython 3.12.16 commands passed, matching every frozen receipt field; all 26 scientific hashes stayed unchanged. Coverage includes exact source/cut reconstruction, whole forward/adjoint and COPY/ownership checks, 11 native controls (including rejection of the untranslated predecessor column), five sets of 512 chart/phase columns, finite physical routes, all 3,038 paid/helper maps and 104 main boundaries, fresh ordinary/owned-stock/grid/envelope/fixed-tape costs, both prices, and 54 strict + 4 functional outer conditions.

git fetch --no-tags https://github.com/CrocSwap/integer-mult-bounds.git main
python3.12 -B scripts/research_package.py --package research/p104-endpoint-temporal-composition

Exact inputs: #7143 9ff298bd6547299dc5b263873bf8a400909e5501 and completed #7142 316b93f25361c0fd1b3a50e1180bd59096c3e45e. Both complete packages are authenticated by inventory, Git blob and SHA256. Five #7142 implementation files remain byte-identical under REUSE.json. Combined native SHA256: 8f5c08866c50070c8579d0dc230bc2941ec3669d643ba373ac4b1d59b423446d. PROOF.md, APPLICATION.md, VALIDATION.md and REPLAY.txt give exact scope and evidence. The separate #7142 receipt is a reported benchmark and source artifact; this package claims full replay of its own combined application, not every separate peer application command.

Credit Rohan Arun for endpoint funding and the signed caller/recharts; Ali Romman for helper and causal-history split identities; Jerry Chen/CubeJerry, Matarener, ToikY, dillona's #7143 source, Cinna Davis/1337isnot1337, Dugongue, DreamingOfClouds, Douglas Colkitt/#7000 reviewers, and every retained author. Substantial OpenAI Codex assistance; no subagents. Apache-2.0.

Five inherited assumptions (native_all_columns, owned_erasure, compiler, precision_tape, analytic_recovery), independent human/formal acceptance, maintainer admission and compiled top-level execution remain open. The selected frontier is unchanged. Hosted scientific CI is separate from local replay. This is one exact composition; its invoice crossing and BIT headroom are recorded, with no global-optimality claim.

Task p104-side-correction: changed source and translated endpoint boundary, with full-cost acceptance. Canonical main e545235, priority revision8 and selected#7000 remained unchanged at the00:52UTC fetch. Full overlap:443openPRs00:46:32UTC and7056Discussionrecords00:47:34UTC; sparse local coverage13working notes/10178HEAD paths,10167unsearched. The final newest-eight PR check found separate zero-gain #7144; exact #7142 remained316b93f25361c0fd1b3a50e1180bd59096c3e45e. No matching composition or stronger complete claim was found within that coverage.

Open next questions: independently replay the translated vector and paid input conversion, then the whole combined bill; refresh guidance/overlap and freeze this source before another optimization.

Canonical investigation and predecessor trail.

Hosted status at public head db8f31c: run42935374695 failed in onboarding:760tests,onefailure,24skipped. CurrentCiNavigationTests.test_current_ci_section_matches_selected_record expects make frontier7000-verify; unchanged CI navigation still references#6030. Scientific replay was skipped. Direct and registered local scientific replays passed; no hosted scientific PASS is claimed.