ci: refresh open PR branches and reconcile shared benchmark changes #426

Problem

Shared benchmark calibration changes remain visible in PRs #399, #400, and #414 after the patch reached main through #401. PRs #414–#423 form a dependency stack and share one inherited calibration commit. Updating them requires preserving each PR's intended base and changes.

Scope and checklist

• Record current PR heads, bases, clean canonical worktrees, and the main commit used for refresh.
• Merge main into the existing branches for fix: key lifecycle state by Click context #399, security: centralize secret-name redaction heuristics #400, and ci: cover consumer fixtures in style and strict typing gates #414 without rewriting history.
• Propagate the refreshed ci: cover consumer fixtures in style and strict typing gates #414 branch into fix: preserve consumer logging ownership and routing #415, then through perf: reuse logging locks and cache source paths #416–release: prepare 0.5.0 candidate and migration guidance #423 in dependency order; preserve PR bases and release: prepare 0.5.0 candidate and migration guidance #423's draft status.
• Check feat: record telemetry span outcomes #402 independently and refresh only if needed for integration.
• Verify every resulting diff against its actual base: remove redundant calibration changes while preserving each PR's intended work, including ci: gate concurrent retention and logging throughput #422's benchmark additions.
• Run appropriate local validation and required hosted checks on the final pushed heads.
• Update PR descriptions with accurate dependencies and validation; record final commit IDs and evidence here.

Acceptance criteria

The redundant calibration patch is absent from the affected review diffs. Intended changes remain intact. Updated branches are pushed, required hosted checks pass, and exact final heads are recorded. No force-push is needed.

This maintenance issue updates existing PR branches; it does not authorize merging feature PRs or publishing a release. The dirty primary checkout must remain untouched.

Prevention

Future shared CI or benchmark-policy changes should use a dedicated prerequisite PR and deliberate dependency propagation. The calibration policy itself is not being changed by this maintenance work.

Execution evidence

Refresh base: a576cc279739eae5e4cfc33ffab2a7fb56de24de. All original branch heads remain ancestors of the updated heads. The primary checkout and its staged changes are unchanged. The #415 documentation conflict preserved both logger ownership and timestamp configuration sections. #402 already contained the refresh base and needed no new commit.

Patch comparison confirms #399/#400/#414 retain their original non-calibration changes; #417–#423 retain identical patch IDs against their actual parents. #415/#416 retain the intended logging changes alongside the main-branch updates. In particular, #422 retains its stress benchmark additions.

Local integrated validation on #423 passed 661 tests and 294 subtests with 86.11% coverage, plus strict typing, style, contracts, benchmark budgets, static security, dependency audit, and uv lock freshness. Every other final head passed lock freshness, baseline, runtime, typing, style, and contracts.

Hosted readback: 2026-10-04T14:19:40.760978+00:00. Required contexts are read from the active default-branch ruleset and checked at each exact PR head; check runs and commit statuses are both included.

Diagnostic reruns

• fix: key lifecycle state by Click context #399: WSL cold import p95 1009.24 ms versus unchanged 1000 ms cap; runtime 610 tests and 262 subtests passed. Same-head rerun success: cold-import median 729.47 ms, p95 753.46 ms; 610 tests and 262 subtests passed (https://github.com/basefoundry/base-cli/actions/runs/37206320662/job/111449668150). Original job 111448140479. No benchmark budget was changed.
• perf: skip contended retention housekeeping #418: Windows persistence median 111.19 ms, p95 714.88 ms versus unchanged 250 ms p95 cap. Same-head rerun success: 2026-10-04T14:08:21.7536274Z base-cli persistence_enabled_ms: median=73.80 ms p95=91.30 ms max=146.52 ms MAD=8.83 ms (https://github.com/basefoundry/base-cli/actions/runs/37206449541/job/111453344999). Original job 111448520689. No benchmark budget was changed.
• fix: preserve consumer logging ownership and routing #415: Unix persistence median 16.25 ms, p95 771.62 ms versus unchanged 125 ms p95 cap; pre-refresh head passed. Same-head rerun success: 2026-10-04T14:16:44.6017614Z base-cli persistence_enabled_ms: median=18.92 ms p95=22.18 ms max=24.61 ms MAD=2.87 ms (https://github.com/basefoundry/base-cli/actions/runs/37206386271/job/111451934580). Original job 111448340008. No benchmark budget was changed.
• ci: gate concurrent retention and logging throughput #422: Unix persistence median 15.35 ms, p95 174.23 ms versus unchanged 125 ms p95 cap; pre-refresh head passed. Same-head rerun success: 2026-10-04T14:17:29.8969950Z base-cli persistence_enabled_ms: median=18.56 ms p95=21.56 ms max=21.64 ms MAD=2.82 ms (https://github.com/basefoundry/base-cli/actions/runs/37206536084/job/111454751722). Original job 111448775778. No benchmark budget was changed.

No feature PR was merged, no branch history was rewritten, and no release was published.

Completion

All 91 required checks (7 contexts across 13 exact PR heads) passed. All four diagnostic benchmark reruns passed at unchanged commits and unchanged limits. Remaining non-required check runs at closure: 0. Branch refresh and review-diff reconciliation are complete; feature PRs remain open and #423 remains draft.