[codex] Cap Actions artifact retention at one day

[kkroo]

Summary

• Lowers perf-test artifact retention values to 1 day.
• Aligns the workflow with the Blockcast org artifact/log retention cap.

Why

GitHub Actions artifact storage was driving billing. Perf outputs should expire quickly unless promoted to durable storage.

Validation

• git diff --check
• Ruby YAML parse over .github/workflows

[allyblockcast]

Ally — Consolidated PR Review

Lenses: pr-review-toolkit (code, tests, comments, errors, types) + gstack/review + native-codex.

Looks good. This is a small, mechanical, low-risk change: retention-days: 90 → 1 on all three actions/upload-artifact steps in perf-test.yml (perf, locomo, obs), each of which is immediately followed by a "publish to dashboard" step that promotes results to durable storage (gh-pages). The stated rationale — billing driven by artifact storage, with durable data already persisted elsewhere — checks out against the workflow structure; no artifact upload was missed and none was left at the old 90-day value. The two test-file diffs are whitespace/formatting-only (dedented assert message, added blank line after docstring) with no assertion or logic change.

Suggestions (2)

• [native-codex] .github/workflows/perf-test.yml — 1-day retention is tight against the workflow's own schedule: "0 6 * * *" cadence. If a scheduled run fails and nobody looks at the raw artifact within ~24h (e.g. over a weekend), it's gone before the publish-to-dashboard step's summarized data can be cross-checked against the raw output. Worth confirming the dashboard capture is sufficient for post-mortems, or bumping to 2-3 days if raw-artifact debugging is ever needed.
• [pr-review-toolkit] hindsight-api-slim/tests/test_document_tracking.py, test_document_id_derivation.py — the formatting-only changes (assert message reflow, blank line) are unrelated to the retention-cap goal stated in the PR description. Harmless, but worth folding into a separate formatting/lint commit next time to keep infra PRs scoped to their stated change.

Strengths

• All three artifact upload sites were updated consistently — no partial application.
• Each lowered-retention artifact has a durable-storage publish step immediately after it, so the tradeoff (short-lived raw artifact vs. long-lived summarized dashboard data) is already covered by existing workflow structure.