[JacobStephens2]

Summary

Keep each Scoring sample attached to the exact file loaded, including arbitrary filenames and samples sharing a seed/window. Resumed model grading, human Scores and Bug-origin settlement all commit through that retained handle.

src/scoring/
├── store.rs                  # decoding, retained identity, private packet lifetime,
│                             # no-clobber publication and atomic replacement
└── origins/settlement.rs     # save decision before withdrawing old Eval proposals
src/scoring.rs                # sampling, grading, human entry and blind views

Preserve the JSON format, ordering, machine-wide writer lock and CLI/Console behavior. Update the domain-to-code index; the inspected hand-written site descriptions remain accurate.

Closes #1305

Evidence

• Before: cargo test --test bug_origins historical_review_and_reclosed_origin_regressions failed in the new save-failure scenario: failed settlement withdrew the retained decision's Eval proposal.
  After: the same scenario passes with withdrawal after successful checkpoint replacement.
• Storage interface: 4 tests passed, covering exact-file ownership without re-enumeration, permissions and payload preservation, same-second publication, no-clobber cleanup, legacy defaults and ordering.
• cargo test --test scoring: 35 passed; cargo test --test bug_origins: 16 passed (each has one ignored subprocess entry point).
• cargo test --test console scoring: 3 passed, including CLI/Console parity with a noncanonical sample filename.
• cargo check --tests, cargo fmt --check, and cargo clippy --all-targets -- -D warnings: passed.
• Final validation: /usr/bin/time -p scripts/affected-tests main passed, using the full cargo test suite because nextest is unavailable. Wall time: 1366.91s (22m 47s), including 40.83s compilation of the remaining test targets; focused builds had already populated dependencies. Timing includes the selector and reflects this host's load, not CI queue time.

Test seams

• Private Scoring storage interface using real temporary filesystem roots.
• Existing Scoring CLI resume and all-five human-entry scenarios, with renamed retained files.
• Existing Bug-origin CLI settlement/count/reclosed scenarios, including renamed files and failed checkpoint replacement.
• Existing shared CLI/Console Scoring actions and blind projections, using the noncanonical filename fixture.

Review coverage

Independent Standards and Spec reviews used git diff main...HEAD, fixed at main a2dfcfbd849ea803eb0e345a2348c6e6a9a81f4f. Both axes read all 11 changed files. Unread changed files: none. Generated-output exclusions: none.

Final reports, including each axis's supporting files-read list, are retained in this Session at .thirdshift-review-PnlyQx/standards.md and .thirdshift-review-PnlyQx/spec.md.

Unaddressed findings

Standards

• Optional Duplicated Code suggestion, declined: tests/scoring.rs:275 and tests/bug_origins/settlement.rs:190 both discover and rename a fixture sample. The Scoring helper serves its two regressions; Bug-origin setup belongs to a separate fixture. Spec decision 12 keeps regressions at the existing CLI seams. Keeping these short setups local avoids adding Scoring-directory knowledge to generic shared test support; the reviewer explicitly calls this tradeoff defensible. No incorrect behavior was alleged. This is a local maintenance decision, with no deferred Spec or Day-shift call.

Spec

None.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Private storage ownership refactor with no JSON migration or CLI changes.

Blast Radius: Scoring

Sample publication and all three writers share this protocol; the existing writer lock and permission-preserving replacement remain in use.

Built with codex · gpt-6.1-sol · xhigh