[addisdev]

Summary

Two assertions in hooks/tests/run.sh's "lanes observe" block required lane_wait_s to be exactly 0. They failed once on a GitHub-hosted macOS runner with 1 (run 40331626282 on #120, a docs-only PR) and passed on a rerun.

Why it flakes: lane_wait_s is WAITED at the moment the lanes first said yes. WAITED starts counting before the hook runs admit_resolve_wait_owner and admit_resolve_worker_log. On a loaded runner that setup can cross a whole-second boundary before the first look, so the first look records 1.

Why not just allow ≤ 1: that would weaken the second check. The bug it guards against is lane_wait_s being re-taken at admission instead of at the first look, and then it would equal waited_s, which can itself be 1.

What it checks instead. The hook writes the held row once, at the end of its first pass, which gives checks that don't depend on timing:

Row	Was	Now
held	lane_wait_s == 0	lane_wait_s is a number, not null, so it was recorded at the first look
admitted	lane_wait_s == 0	equals the held row's lane_wait_s, and lane_wait_s < waited_s (not re-taken at admission, which is always at least one poll later)

This is the same idea as #76, which times two other "promptly" checks from the hook's own events. The two PRs touch different hunks of run.sh.

Mutation-checked: I temporarily made the hook overwrite ADMIT_LANE_WAIT on every pass (dropping the [ -n "$ADMIT_LANE_WAIT" ] || guard). Both admitted-row checks fail (expected '0', got '2' and not before). The hook was restored and is not part of this diff.

Second commit (91f53b7): the first push put a case … esac inside $(…). macOS's /bin/bash 3.2 can't parse that, so CI stopped at line 972 with a syntax error. Bash 5 and bash -n on Linux accept it. The case now sets a variable that the ok line compares; no other $(case exists in the repo.

Platform tested on

• macOS: not run locally on macOS. CI's Shell tests job runs this file on macos-latest, and that's what caught the bash 3.2 parse error above. I couldn't get a bash 3.2 binary into this environment (GNU's FTP site and GitHub mirrors are blocked by its proxy), so macOS CI is the bash 3.2 check.
• Linux (cloud container, bash 5.2): the whole lanes-observe block passes. Two other checks, "beta got in promptly" and "orphaned hook exited", fail in this container on unmodified main too (148 pass / 2 fail on main, 149 / 2 with this change). So they're environmental here and unrelated.
• Node: node -v → v22.22.0 (not used by this test)

Checklist

• Tests pass: cd dashboard && npm test (no dashboard change)
• Shell syntax clean: bash -n OK; shellcheck --severity=warning hooks/tests/run.sh clean (same 5 info-level notes as main)
• No host-specific data: CI's "Release check (offline)" passed on the first push; couldn't run locally (no authenticated gh)
• Docs updated if behaviour changed (test-only change)
• Schema migration (n/a)
• New environment variables (n/a)
• Security-sensitive change (n/a)

🤖 Generated with Claude Code

https://claude.ai/code/session_019W86FcknjPBqAPtVcnH8Ja