[addisdev]

Summary

The README says the queue classifier has eight causes and that exactly one offers to add a runner. Neither matches the code:

• dashboard/lib/queue-cause.js defines ten entries in CAUSES.
• Three of them return actionEligible = true: unserved, role-unserved and repo-capacity.

docs/concepts.md (the README's own link target) and docs/figures/queue-causes.html already say ten and three. Only the README was behind. tayloraddison.dev's case study on this project cites ten, which is how the mismatch came up.

Two comments in queue-cause.js carried older counts too, so I fixed them in the same change:

Was	Now
"That answers two of seven causes"	ten
"for the six causes marked false"	seven are marked false
The unserved branch called itself action-eligible "unlike every other non-capacity cause"	not true since role-unserved exists; now names both

Comments and README only, no behaviour change.

Platform tested on

• macOS: not tested locally on macOS; the macOS CI jobs (Preflight smoke test, Fleet Cockpit, Shell tests) pass. Nothing here is platform-specific.
• Node: node -v → v22.22.0

Checklist

• Tests pass: cd dashboard && npm test (1112 tests: 1111 pass, 0 fail locally; green in CI)
• Shell syntax clean: no shell files changed; CI "Shell checks" green
• No host-specific data: scripts/release-check.sh --offline. It couldn't run locally (no authenticated gh), but CI's "Release check (offline)" passed on a43fa05.
• Docs updated if behaviour changed (no behaviour change; README brought in line with docs/concepts.md)
• Schema migration (n/a)
• New environment variables (n/a)
• Security-sensitive change (n/a)

Also run: node --check on every tracked .js file, make-skill.mjs --check ("skill is up to date"), make-glance-fixtures.mjs --check (exit 0) and scripts/check-docs.sh (done, all ok).

CI note: "Shell tests" failed once on two lane_wait_s == 0 timing assertions in the admission-hook tests, which this PR doesn't touch (details in the comment below). It passed on one rerun.

🤖 Generated with Claude Code

https://claude.ai/code/session_019W86FcknjPBqAPtVcnH8Ja

[addisdev]

Shell tests failed on a43fa05, in two assertions of hooks/tests/run.sh's "lanes observe" block:

• ...from the start of the wait (line 965): expected 0, got 1
• the lanes would have let it in at once (line 978): expected 0, got 1

Everything else passed (148 of 150 in that file), and so did the other eight jobs.

This PR changes only a README paragraph and three comments in dashboard/lib/queue-cause.js, nothing the admission hook runs. Both assertions require lane_wait_s, a whole-second count, to be exactly 0. On a loaded GitHub-hosted macOS runner the hook can cross a second boundary between its start and its decision, which gives 1. main passed the same job on the same base (9a87c00, run 40129594328).

There's no fix for these two yet. #76 makes a similar change for two other timing checks but doesn't touch the lanes-observe ones. A patch in that spirit would be to allow a one-second tolerance, e.g. [ "$(field … lane_wait_s)" -le 1 ], since these checks exist to catch a lane that held the job, not a sub-second tick. I haven't widened this PR to include it.

I'm rerunning the failed job once.

---

Generated by Claude Code

[addisdev]

The lane_wait_s timing flake from the earlier failure now has its own fix in #121. Instead of a 1-second slack, it checks the first look by structure, so it can still catch a lane wait re-taken at admission. This PR doesn't need it, since it's green on the rerun.

---

Generated by Claude Code