[JacobStephens2]

Summary

metrics read
  local Scoring checkpoints → latest decision per bug
  confirmed settlement → one retained Run
    local attribution: exact Command record + Run name
    cross-host attribution: latest local Run for the settled PR
  Run details + that Run's start-week trends → confirmed_bug_origins

Rejected and unsettled Bug origins count for nothing. Late confirmations and corrections are joined when metrics are read; old Command records remain unchanged. The manual and hand-written site pages describe the behavior.

Closes #1039

Evidence

• Before: cargo test --test metrics confirmed_bug_origins_count_at_read_time_against_the_run_and_its_week failed with a missing count (null, expected 0). The Spec review's exact continuation reproduction failed: one confirmed Bug origin counted 2.0 times.
  After: the same test and exact reproduction pass. cargo test --test metrics passes 29 tests (one subprocess fixture ignored), including late confirmation, corrections, rejection, reopened bugs, child Runs, retained metrics and continuation attribution. cargo check passed during implementation.
• Local validation: scripts/affected-tests issue-981 ran for 959.27s, including compilation. All 44 site tests and 2,250 Rust tests passed before site_deploy stopped on an unrelated host snapd.service warning (Unknown key name 'RestartMode'). SYSTEMD_LOG_LEVEL=err cargo test --test site_deploy the_systemd_units_verify passes. The affected run used process-only Git configuration to disable this host's global commit/tag signing defaults; machine settings were unchanged.
• Full CI on head c0dd9930: passed, including the full Linux Rust suite, formatting, clippy, Python tooling, site and Console browser checks. Exact-head CI run. Release build and smoke checks also passed.

Test seams

• The whole binary against the existing Scenario: retained Command records and Scoring checkpoints go in; thirdshift metrics <command> --json, text output, and weekly trends come out.

Review

Fixed point: issue-981 (c77938b014d94c68438abbf569e1ba55f81f2608). Independent Standards and Spec reviewers inspected eleven of the twelve final changed files, including generated Cargo.lock and site/docs/index.html; no generated-output exclusions.

Changed file left unread by both independent review axes: tests/run_metrics.rs. It changed after their follow-ups to add the new read-time confirmed_bug_origins: 0 field to two existing whole-object expectations, while continuing to compare every retained metric. The implementer inspected it; cargo test --test run_metrics passes 10 tests (one subprocess fixture ignored). Each report explicitly records this final coverage gap.

• Standards: 0 findings.
• Spec: 1 finding, fixed. A confirmed Bug origin was duplicated across continuation Runs sharing a PR. The exact failing reproduction was run before deciding, then passed after the fix; the regression remains in tests/metrics.rs.

Full reports, including each axis's final files-read list, are saved with this Session at .thirdshift-review-DGsqRr/standards.md and .thirdshift-review-DGsqRr/spec.md.

Unaddressed findings

Standards

None.

Spec

None.

Merge Danger

Door: two-way

The change only reads local Scoring data and adds a metrics signal; it does not rewrite retained records.

Blast Radius: metrics

Malformed Scoring checkpoints report an error instead of silently showing zero Bug origins. Attribution follows the Scoring checkpoint format already on main; the specified base branch does not yet contain that writer.

Built with codex · gpt-6.1-sol · high