Signals & Insights

Healthy

Rate is above 70–80%. Objectives are well-scoped, acceptance criteria are explicit, and agents execute against clear expectations without needing a second attempt.

Watch for

Rate below 60% or declining. Almost always a direction problem — objectives are ambiguous, over-scoped, or the definition of done isn't explicit enough. Audit rejected objectives before adjusting agent configuration.

Healthy

Objectives complete in hours to a few days, consistently across similar work. Cycle time is predictable and doesn't surprise the team.

Watch for

Cycle times drifting toward weeks, or high variance between similar objectives. Usually signals objectives are too broad, or a systemic delay is accumulating at one handoff stage.

Healthy

Pickup is short — objectives are claimed quickly after scheduling. Execution is proportional to complexity. Review is fast — humans are not the bottleneck.

Watch for

Long pickup means agents aren't running frequently enough. Long execution means mid-run ambiguity or scope expansion. Long review is the most common agentic bottleneck: agents ship in hours, but throughput is capped by how fast humans approve.

Healthy

Rate is high and stable, or rising. Agents execute cleanly against well-written objectives. Human direction quality compounds: each well-scoped objective makes the next easier to write.

Watch for

Rate is low or declining. Almost always an objective quality issue, not an agent issue. Ask: were acceptance criteria explicit? Did the objective describe an outcome or an implementation?

Healthy

Rate is stable or improving across consecutive windows. The team is getting more deliberate about defining objectives before scheduling them.

Watch for

Rate declining across multiple windows — the team is scheduling objectives that aren't ready. Usually a sign that urgency is overriding definition quality.

Healthy

Rates are consistent across agents for the same class of work, or the team has learned which agents work best for which objective types and routes accordingly.

Watch for

One agent kind consistently underperforming on a class of work. Usually means those objectives need more context for that agent — file pointers, interface contracts, explicit criteria — not a different agent.

Healthy

Clarity flags are rare. When raised, the team refines the objective before running. First-pass rates on flagged objectives, once resolved, are comparable to unflagged ones.

Watch for

High flag rate, or flags consistently overridden and objectives run anyway. An overridden clarity flag is the strongest predictor of rejection — the agent couldn't interpret the objective, so the first attempt is almost certain to miss.

Healthy

A small number of directors producing a large and growing volume of agent execution. The ratio improves over time as objective quality increases and agents run more efficiently.

Watch for

Low or flat leverage with multiple directors active. Either objectives are cycling through rework (direction quality), or agents aren't being run frequently enough (utilization). Stage latency helps tell the two apart.

Healthy

Multiple objectives flowing in parallel — agents executing while others are in review or pickup. A team of two or three directors can sustainably run three to five objectives concurrently.

Watch for

Concurrency consistently at 1. The team is running objectives sequentially when agents have capacity for more. Increasing concurrency is almost always the fastest path to higher throughput without adding people.

Healthy

The distribution of delivered work broadly matches what was planned. The team follows through on stated strategic priorities rather than defaulting to whatever is easiest to execute.

Watch for

Persistent gaps between planned and delivered for a specific bucket — most commonly TRUST work that's always planned but rarely shipped. This pattern is invisible in task-based tools and immediately visible here.

Healthy

Coverage is 90% or higher. Agents classify objectives at claim time automatically, so gaps are rare and limited to early objectives created before the workflow was established.

Watch for

Coverage below 80%. The portfolio allocation view is incomplete and unreliable — the team is making priority decisions without a full picture of where effort is actually going.

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.

Healthy

Chain passes verification. Every record hashes correctly to the previous one and signatures verify against the workspace signing key in KMS. The audit record is tamper-evident and satisfies the 2026 AICPA SOC 2 criteria for AI-generated code.

Watch for

Chain fails verification. May indicate data tampering or a system error — either requires immediate investigation. The integrity of the compliance record is at stake: a broken chain cannot be used to demonstrate a verifiable chain of custody to an auditor.