Repository navigation
📊 Refresh runtime test weights from measurement 74 - #1110
Conversation
PR #1110: 📊 Refresh runtime test weights from measurement 741 files, +1378 / -1351 Scope🔴 PR has 2729 lines changed. Split into focused PRs. 🟡 2729 lines changed. PRs under 400 receive more thorough review. Structural✅ No structural bloat detected. Slop✅ Slop indicators look low. Static Analysis✅ Oxlint found no issues. CorrectnessNo extraneous code patterns detected. |
taras
left a comment
taras
left a comment
There was a problem hiding this comment.
Independent Codex Astra (gpt-6-astra, high) review PASS at 8d552b5c7e636bc4a5a5cf51b9da1f5f8145b180, replacing the required review after a joined Fable quota refusal. Original failed invocation retained; no source or counter reset.
W1–W3 pass for the authorized data-only import. Only test-weights.json changes from the specified base; committed bytes match the archived artifact and required SHA256. Reused signature and focused-test receipts (10 groups, 52 steps); independently confirmed complete runtime records, exact provenance, unique coverage and every timing. Later corpus drift and conservative fallback are explicit. The Bun calibration incompatibility requires separate disposition, as the original request directs. Source remains unchanged.
- Separate calibration disposition: test-weights.json:991 records Bun adversarial-implementation.test.ts at 421949 ms, exceeding the 420000 ms target by itself. scripts/lib/test-shards.ts:64 applies that same fallback to the later-added repl-unfinished-entry.test.ts. The resulting applicable total is 3853981 ms, giving the documented floor of 10, while .github/workflows/ci.yml:656 retains 9 shards. Increasing counts alone cannot resolve the oversized individual file. specs/testing-spec.md:855–861 requires reporting this incompatibility; the original request expressly prohibits count changes or new calibration within this import. These facts do not invalidate the artifact or partition coverage, but the passing focused regressions do not establish seven-minute calibration. Carry this explicit gap into the coordinator's separate calibration disposition.
The artifact import is exact and valid. This verdict establishes no seven-minute calibration claim and changes no shard count. Required CI and the new PR merge decision remain separate.
Refresh the CI partition weights with the exact assembled artifact from measurement74, attempt1. It measured integrated main
c128b655on ubuntu-latest with Deno2.9.5, Node22.23.4 and Bun1.4.0. All three runtime jobs and assembly succeeded.Only
test-weights.jsonchanges. Its bytes match artifact13336906926 exactly (SHA256a35633308f9748b03147265009036d9a3c9655e5a2c93481c6d67fc3667dc957); all ZIP digests, complete records, provenance, unique coverage and every timing were cross-checked. No timing is edited or stitched from cancelled run73.Focused verification passes:
deno task test scripts/tests/test-weights.test.ts scripts/tests/test-shards.test.ts scripts/tests/measure-test-weights.test.ts—10groups/52steps, zero failures. Canonical parse/format and artifact/worktree/commit equality pass. The import commit is normally signed.Current main has one later-added file,
packages/cli/tests/repl-unfinished-entry.test.ts, which uses the existing conservative fallback. Two measured files changed after the captured revision:repl-sidekick-controls.test.tsandprotected-input.test.ts; their recorded timings retain their actual captured provenance.Shard matrices remain19/16/9. The current Bun corpus plus its unmeasured-file fallback has a documented theoretical floor of10; its largest recorded file is421949ms. The independent Astra review passes the authorized exact-artifact import; the separate calibration incompatibility is tracked in #1111. A successful correctness CI run alone does not establish the separate seven-minute calibration claim. No shard count, runtime source, dependency, exclusion or qualification scenario changes here.
This follows #1100. The unrelated missing-diagnostic investigation #1107 remains open; this artifact does not establish its cause or repair.