Keep viable research PRs active through repair and harvest

[vantasnerdan]

Advances #81

Outcome

This corrects the process failure exposed by the premature closure of #77 and removes unconditional full-suite validation from ordinary PRs.

• refactor then merge requires an owner or handoff, live source/harvest PR, exact repair, and landing test.
• Unmerged closure is limited to enumerated terminal cases with unit-level evidence and landed replacement links.
• Incomplete accepted dependency closure, canon conflict, unavailable self-merge authority, and pending finite repair are explicitly non-terminal.
• Accepted canon remains authoritative for releases and promotion but scientifically challengeable; correct conditional artifacts can merge without promotion.
• PR validation now chooses full, affected-test scope, or fixed checks only from the base diff. Additive new modules with matching tests may stay scoped; existing framework-module changes, governance/shared machinery, removals, untested or multi-module additions, and uncertain executable changes force full.
• Full integrated validation runs weekly or on manual request, rather than on every PR and again after every merge.

The root contract, onboarding, PR template, research harvest and physics skills, five memory templates, workflow/validation scripts, and regression tests agree.

Scope and authority

• Scientific claims promoted: none.
• Accepted claims/releases changed: none.
• Goal [Process] Keep viable scientific PRs open and actively harvest or challenge canon #81 complete locally: yes; delivery remains pending independent review/merge.
• Related scientific lifecycle: source PR Exact-mass one-loop proper-time coefficients with preregistered regulator comparison (P230, advances #76) #77 remains open while focused harvest Harvest exact-mass one-loop regulator atoms #82 is reviewed.

Validation boundary

• scripts/validate_changed.py --base main --head HEAD --print-only selects the four affected test files below.
• scripts/validate.sh --pytest-scope tests/test_public_contribution_surfaces.py tests/test_repository_validation.py tests/test_validate_changed.py tests/test_validate_script.py — PASS: all fixed repository checks plus 32 focused tests.
• Conservative selector tests cover full/scoped/fixed-only dispatch, existing/new modules, additive/removal public exports, governance, removed tests, workflow drivers, and unmapped documentation.
• Physics skill validator and skill-creator quick validation — PASS.
• GitNexus final change detection — LOW risk, zero affected processes.
• git diff --check — PASS separately.

No full pytest replay was used for this boundary. The earlier interrupted local full attempt and the workflow run produced before the CI correction are not the evidence for this revision.

Merge authority

I authored and implemented this PR and will not merge it. A distinct reviewing agent or repository owner must decide and perform the merge.

[vantasnerdan]

CI verification after the workflow correction: Actions run 32122559445 completed successfully in 54 seconds. The log reports Validation mode: scoped, selects exactly the four declared test files, runs 32 tests, and ends with ALL FIXED REPOSITORY CHECKS AND REQUESTED PYTEST SCOPE PASS. The periodic/manual full-validation step was skipped for this PR as intended.

[axis-marbell]

Administrative CODEOWNER approval on maintainer authority; no code review performed.