Skip to content

Human action 22

Jerry Holland edited this page Sep 29, 2026 · 3 revisions

Record the identified Windows benchmark baseline

Canonical work item: #22.

Completed 2026-09-09. lab-windows-a was recorded on the quiet Windows host and qualifies: reference hosts.lab-windows-a, latencySkipped false, 30 comparisons passing and none skipped. Evidence is on the issue. This procedure stays in force for the next re-record; substitute the assigned label wherever the steps below name one, and note that the expected publication count is now 42 rather than the 41 these steps were written against.

When required: Before Windows latency qualification. The issue carries current status, decisions, and completion evidence.

Owner role: Windows qualification-machine owner.

Prerequisites: Quiet Windows host, clean current checkout, stable public-safe host label, and no competing suites. Use intended platform defaults without engine/session overrides for historical deterministic comparison.

  1. Run npm run test:benchmark -- --host <assigned-windows-label> before recording. An initial latency skip is expected for an unknown identity; inspect the deterministic findings.
  2. Confirm the expected publication count is 42 versus the preserved historical 41. Investigate any other growth or unexpected count before recording.
  3. Run npm run benchmark:record -- --host <assigned-windows-label> deliberately on the quiet machine, then npm run test:benchmark -- --host <assigned-windows-label> --json.
  4. Confirm the selected reference is hosts.<assigned-windows-label>, latencySkipped is false, and comparisons pass. Review CPU/memory, architecture, OS/toolchain, overrides, and the baseline diff.
  5. Commit the reviewed baseline with its explanation and retain validation evidence on the follow-up. Coordinate closure of the original #22 only after its acceptance criteria are met.

Verification: An identified baseline and passing matching-host check must exist. Preserve other entries and historical measurements; do not relax tolerances.

Failure handling and cleanup: On unexpected growth, preserve output and investigate before re-recording. Benchmark fixtures normally clean themselves; remove only identified disposable leftovers after evidence is retained.

Completion: attach the evidence/decision to issue #22, resolve its human-action label when satisfied, update the private board, and point the Human TODO entry to the completed outcome.

Publish only evidence suitable for the public issue/wiki. Redact private paths, repository content, and credentials from shared output. Sensitive vulnerability findings belong in a draft security advisory.

Back to Human TODO · Benchmark host baselines

Clone this wiki locally