free-durations-record.yml runs Mondays 04:00 UTC (before test-health) and on
dispatch. Linux records on ubicloud-standard-8, the free-suite shard runner,
with scripts/ubicloud/setup-free-suite.sh; Windows records through
windows-free-tests.yml's record_durations path (now also workflow_call).
When either seed's file set drifts from main, the propose job force-updates
bot/free-duration-seeds, opens or refreshes one seed PR stating the seed
ratchet result and the files that failed while recording, and dispatches the
free lanes on it. A red file keeps its duration; an interrupted recording
uploads no seed (this also stops the Windows dispatch from uploading a stale
copy), and more than 5 red files on either OS proposes nothing. The bot never
edits the allowlist. open_pr=false on dispatch records and reports only.
The workflow is classified free-lane only in scripts/test-pr-profile.ts, with
a non-behavioral HARNESS_VERSION decision.
Intentional pin change: test/windows-native-workflows.test.ts now expects
workflow_call and the complete-recording guard on the Windows record job.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
