test hygiene: desktop-notification C1 slow-sink threshold margin thin on Windows

[kyle-sexton]

Observed during #443: desktop-notification.test.sh's C1 fd1-leak detector asserts the hook returns in < 2000ms against a 3s-sleeping sink. After the hook::buffer_stdin migration the hook's wall time on a Windows Git Bash dev machine is ~1.6s (process-spawn overhead: the helper's jq -e . probe plus tr/awk subshells), leaving ~400ms margin. Under concurrent load (parallel test suites, or the suite itself run inside a command substitution) the assertion false-fails with 4-10s readings even though no fd1 leak exists — verified by an isolated run (1595ms PASS) and a main-baseline (896ms).

Ubuntu CI is unaffected (spawns ~10x cheaper). Options when this bites:

• Assert leak vs no-leak by comparing elapsed against the sink's sleep (e.g. < sleep_duration) instead of a fixed 2000ms — the actual invariant is "did not wait for the sink".
• Or bump the threshold and the sink sleep together (e.g. sleep 6 / threshold 4000) to widen margin proportionally.

Not urgent; file so the next person hitting a local slow-sink FAIL on Windows finds this instead of re-diagnosing.

Related

#443 (migration that thinned the margin). Epic #313 (closed).

No linked issue: test-hygiene note; closes nothing.

[kyle-sexton]

This was generated by AI during triage (historic normalization, rule 8g).

Both options in the issue body are viable; picking the one that fixes the actual invariant rather than papering over the symptom.

Decision defaulted: assert elapsed time against the sink's own sleep duration (e.g. < sleep_duration) instead of a fixed 2000ms threshold — veto before merge (this is the real invariant — "did not wait for the sink" — and stays correct regardless of machine/shell spawn overhead; the alternative, bumping both threshold and sink sleep proportionally, only buys margin at the current overhead level and would need re-tuning again on a different machine).