[petalas]

Summary

Add a dedicated scripts/find_source_overlap.py CLI for source references, PR authors and creation dates. This is workflow tooling with behavioral regression evidence, produced by Nick Petalas. No mathematical result, kappa improvement or verification of a predecessor result is claimed.

--reference NUMBER       identifier-aware source lookup, repeatable
--pr-author LOGIN        restrict the PR body corpus to one author
--created-since DATE     inclusive UTC creation cutoff

The new CLI reuses the existing normalizer, identifier matcher, PR deduplication and atomic cache writer without modifying or monkeypatching them. scripts/find_overlap.py is byte-identical to the PR base. The dedicated PR-only cache loader retains author and creation metadata, refreshes legacy metadata when needed, and reports unavailable filter metadata explicitly. Closed/merged bodies require --include-closed and have separate coverage timestamps. Local notes and Discussions are outside this intake command.

Evidence

Eleven new intake regression groups and all 37 existing overlap-named groups pass in normal and optimized Python. Controls cover identifier noise, author/date boundaries, legacy caches, optional closed bodies, malformed input, failed cache publication, and unchanged legacy function/record behavior. The standalone CLI also retrieves PR6571 as a source-6537 lead from the retained 587-open-PR snapshot. This is retrieval evidence only; no mathematics in either PR was evaluated.

The registered package, generated input-hash evidence, AST/function-size checks, whitespace, and exact committed-byte/source-history check pass at b994ea2487733dace57c67bed957bb95bf3fc745. Both registered checks finish in about two seconds locally, below their 120-second limits. These are check-duration observations, not a speedup claim.

From the repository root:

/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/source-reference-intake
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -B research/source-reference-intake/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/source-reference-intake --check-source-history

For an intake query, substitute the actual source, login and UTC cutoff:

/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/find_source_overlap.py --reference 6537 --pr-author petalas --created-since 2026-10-27 --include-closed --cache /tmp/source-overlap.json

Another host may substitute its CPython 3.12 path. Standard library only.

Limitations and unrun obligations

References cannot detect uncited semantic counterparts or decide novelty. Snapshots are not atomic, and the command is not a policy decision about a proposed target. No mathematical predecessor replay, selected cold replay, physical supplier admission or independent human review was run. No other contributor's package, pinned evidence, frozen source archive, upstream file or selected-result record is modified.

The delivered tooling scope is recorded in the Discussion scope update.

Open next questions

Can a separately scoped package-provenance index expose uncited reuse? First compare exact package-path references with a complete changed-file inventory. Missing citations, renamed packages and semantic equivalence must remain explicit coverage limits.

Related open work

• PR6334 preserves distinct Git paths in the existing overlap helper. This PR leaves that helper byte-identical and supplies a separate PR-source intake command. No common scientific quantity is computed.

Related open work was refreshed after CI. No additional contribution studying this source-intake interface was found among this account's open PRs created since the CLAIM; PR6334 remains open.

CI notes

The initial run 41652301463, at head e55541991331d578d88660dcbb34af60506a534d, passed plan, source-reference-intake, overlap-reference-prefilter, overlap-discussion-batching and replay-input-head-spelling. The overlap-path-index job failed with:

source blob provenance mismatch: scripts/find_overlap.py

This PR caused that failure by changing the shared file and selecting a historical package that pins its blob. Commit b994ea2487733dace57c67bed957bb95bf3fc745 restores the exact PR-base bytes and moves the feature into the dedicated CLI. The other package and its pins were not edited. Local replay planning against the PR base now selects only research/source-reference-intake; the historical package is not selected by this diff. This is an isolation fix, not a claim that the historical replay was rerun successfully.

Both plan and replay (research/source-reference-intake) passed at head b994ea2487733dace57c67bed957bb95bf3fc745 in run 41656250857. The completed check list was read once and confirmed both successful jobs. Hosted replay certified the registered normal and optimized 48-test suite. This is behavioral replay evidence, not independent human or mathematical proof-kernel review.

Merge Danger

Door: two-way. The new command is reversible.

Blast Radius: dedicated intake CLI. The existing helper and all frozen evidence are unchanged.