Board drops a card when the task's only worker row is :completed (e.g. a finished CI fix pass) but the task is still in_progress

[ryanrborn]

Summary

An in_progress task whose only surviving author worker row was :completed (e.g. a finished CI fix pass) was invisible on the board entirely. worked (apps/arbiter/lib/arbiter/board/snapshot.ex) counted every author worker row for a task regardless of status, which stopped orphaned_cards/3 from picking the task up. Since :completed sits in neither @running_statuses nor @waiting_statuses, the task matched none of the board's card-producing filters — even though classify_columns/2 still called it :waiting.

Fix: orphan_worked/2 narrows worked to task ids that have at least one non-:completed author row before it's used by orphaned_cards. A task whose author rows are ALL :completed now falls through to the orphaned-card path and renders a Waiting card. The orphaned card's reason now also names review_park_reason (e.g. review-parked (resume_blocked) — resume or close) when the issue carries one, instead of the generic "worker stopped" message.

Test plan

• New test: an in_progress issue whose only worker row is :completed renders a Waiting card whose reason names review_park_reason.
• New test: same, without a park reason, falls back to the generic "worker stopped" message.
• New test: a task with both a :completed row and a live row is not double-counted.
• New invariant test: every issue classify_columns/2 places in :waiting or :running yields exactly one card from derive/1.
• mix test test/arbiter/board/ and test/arbiter/tasks/epic_rollup_test.exs — 290 tests, 0 failures.
• mix precommit (compile --warnings-as-errors, deps.unlock --unused, format, test) — clean except 6 pre-existing failures (ProvisioningTest ×4, CLI Create/ReleaseDeploy ×2) caused by this worker's own ARB_* environment variables, not by this change; confirmed by re-running ProvisioningTest with those vars unset (0 failures).

References

bd-6lvc1r

Closes #2021

🤖 Generated with Claude Code