Risk signals

Healthy

Flags are rare. When raised, the team treats them as a signal to refine the objective — zero execution time spent, zero code written. The flag resolves the ambiguity before it becomes a rejection.

Watch for

High flag rate, especially if the team is overriding flags and running objectives anyway. A high rate means objectives consistently lack specificity — the fix is explicit acceptance criteria, file pointers, or narrower scope before scheduling.

Healthy

Multi-rejections are rare. Single rejections are expected — the review gate working as intended. Objectives that fail once are resubmitted with tighter scope or clearer criteria and pass on the next attempt.

Watch for

Objectives cycling through three or more rejections. Signals that acceptance criteria aren't explicit enough, the objective is too large and gets partially completed each time, or the expected output isn't achievable in a single run. These should be split, not re-run.

Healthy

Objectives move through lanes in hours to days. Backlog is shallow and well-curated. Scheduled objectives are claimed quickly. In-progress objectives complete before they go stale.

Watch for

Aging backlog is prioritization debt — more work is being created than executed. Aging scheduled means agents aren't running. Aging in-progress means a blocker surfaced mid-run that wasn't anticipated in the original objective.

Healthy

No stale objectives. Agent runs either complete and request acceptance or surface a blocker quickly. The board reflects real-time state — nothing is left sitting indefinitely.

Watch for

Any stale in-progress objective. These represent blocked or abandoned runs, or work waiting on an external dependency that was never surfaced. The right action is to reject and reschedule with fresh context — not leave them in limbo.