[flyingviper]

The selected PR5599 independent caller adapter assumes eight owned lifetimes and one reusable scratch field. Newer P104 words have overlapping lifetimes, so that adapter does not directly cover their interface. This isolated package replaces its numerical inventory assumptions with an explicit independent/owned partition and a complete two-field scratch schedule, then audits PR5713 at a5348f7143f95fe0046b92b96172e30763c7d731.

The unchanged PR5550 positive control reproduces its complete frozen output. The current candidate passes the independent 32-bank free-Laurent caller, signed-phase/normalizer checks, all eighteen forward LOCALs, 66,348 paid jobs, 12,958 targets and sixteen rejection controls. Its eleven local owned lifetimes peak at two simultaneous fields; the global schedule includes thin-query reuse. A separate exact role-permutation fixture creates a retired public hole and checks all 1,758 local/query intervals. The ordinary pass count is 124,284,844, below the retained 184,647,552 allowance.

This closes a source-specific caller evidence gap. It does not introduce a construction, improve kappa, independently replay every paid native map, admit the complete supplier, or change the selected result. The five retained framework assumptions and numerical residuals remain explicit. The PR5713 primary verifier passed separately, including 2,917 native maps/phases and complete pricing; that prerequisite log is distinguished from this package's independent coverage.

Validation on Python 3.12: package metadata/pins; registered normal and optimized replay; committed bytes and source history; official 45-file export and integrity check; and Git-free replay with all exported files marked read-only and hashes unchanged afterward. Every run matched complete scientific output hashes, not just a success marker. A separate agent checked the integrated source binding, literal inventory, interval schedule and receipt consistency. These are independent AI checks, not independent human or formal acceptance.

Reproduce:

python3 -B scripts/research_package.py --package research/p104-caller-owned-interface
python3 -B scripts/check_research_commit.py --package research/p104-caller-owned-interface --check-source-history

Both registered commands run the original and candidate serially, with separate 120-second/512-MiB child caps. Each pair took about 50 seconds locally. Main was refreshed at df0137efd35e11b4445734bf612bdd4c7ebc330c; PR5713's head was unchanged before publication. The open P104/caller PR search and current handoff discussion were checked for overlap; coverage does not include unpublished work.

Credit Chris R/Clear1nsight's PR5550 independent checker, the maintainer's PR5599 adapter, Ali Romman's PR5713 and PR5678, Joel Pulikkan/GamingPuzzled's PR5684, and all retained source attribution. Prepared for flyingviper with substantial OpenAI Codex assistance. Apache-2.0. Frozen predecessor packages and shared framework files are unchanged. Open next questions and remaining admission obligations are in the package README and TASK.json.

CI status: the hosted planning job failed in the unchanged upstream onboarding test test_current_ci_section_matches_selected_record, on the missing review and source_manifest documentation links. The scientific replay job was skipped. The same two failures reproduce locally on unchanged canonical-main documentation; all scoped package checks listed above pass. Existing PR #5703 supplies that documentation fix. Job evidence: https://github.com/CrocSwap/integer-mult-bounds/actions/runs/40314285344/job/121542860148. No tests or CI predicates are weakened by this package.