[dillona]

Three simultaneous contact-free retimings on exact #7112 expose one more independent helper split, rank9 → rank3 + rank6. The conditional claim becomes κ = .003638315828316, up .000000898708375 over source #7112 and .000000336058316 over the concurrent best open claim #7113. The 18-stage signed caller is retained; #7113's independent 17-stage caller is acknowledged but not imported.

Each retimed role preserves its exact paid-rank histogram. The full emitted word adds 686 ADDs and 4 SCALEs; all are charged. Per LOCAL ΔP={} and ΔT={3:+1,6:+1,9:−1}; the full invoice has117,148 paid and23,482 productive requests. Ordinary bound849,950,512<10^9. The unchanged grid is1125·2^24, with freshly recomputed precision. Independent raw bracket is(.00363831582831655596050722548632002,.00363831582831655596050722548632003).

• Exact source head: e59520b (Conditional κ = 0.003637417119941 (+9.6926221e-7 over #7110) #7112); all19 predecessor files and their nested source closure are frozen offline.
• Candidate canonical word SHA256: ae30929ff3b5f6b49a5edac78e2164acfbf809113d0b9f03249ef940dd7b2491.
• Complete local CPython3.14.7 replay passed. The registered isolated CPython3.12.16 replay then matched every frozen receipt, including source reconstruction, arbitrary-input forward/adjoint, exact signed caller, two price engines, all ordinary/precision/compiler/outer gates, 512-column chart phase checks, full finite native/signed-route physical replays and mutation controls.
• Metadata/pins and committed bytes/modes pass. Source-history preflight passes at scientific commit96cf946d9199ff5222a564fc852da5799f31ccca on canonical main e545235.

python3.12 -B scripts/research_package.py --package research/p104-free-cohort-split

Publication commit e63533f carries exactly the scientific package tree c99d054422c82a46f5e1128317a99477c183a2be. It is package-only atop the older account fork because the credential cannot push workflow-history changes; canonical provenance remains unchanged. The standalone replay needs neither Git nor network. Hosted run42862736742 failed before scientific replay: 760 onboarding tests, one failure in test_current_ci_section_matches_selected_record (guide still names6030; selected record requires make frontier7000-verify),24 skips. The scientific job was skipped. This upstream failure is documented in the existing handoff. #7113 separately carries a guide correction.

The five framework assumptions, complete changed-source occurrence/storage/root and physical-upper-cost application, and independent review remain obligations. This is a finite source-bound conditional result, not a selected-frontier update or complete multiplication theorem. No shared code, selected record or frozen predecessor is changed.

Credit: Ali Romman (helper/basis method), Jerry Chen/CubeJerry (retiming, inventories, caller/BIT/precision), DreamingOfClouds (native chain), ToikY (#7003/#7004/#7108/#7109/#7110), the preceding #7105/#7107/#7112 edits, Rohan Arun's concurrent #7113, and all retained notices. Prepared for dillona with substantial OpenAI Codex assistance; no subagents. Exact scope, proof, evidence, ceilings/headroom and Open next questions are in the package README and TASK.json. Canonical investigation.