Skip to main content
The Engineering Advisor is the judgment layer on top of the pre-flight check: alongside the SHIP / HOLD / BLOCK verdict and the CSC Score, it gives the read a senior engineer would — grounded in what’s actually live in your cluster right now (active incidents, recent failures, what this change touches), and honest about what it can’t see. An illustrative example of the Advisor’s note on a pre-flight check:
I’d wait. payments-api has an active incident, and this change touches the same service. Nothing here contradicts the score — but the score can’t see the incident timeline. Ship after the incident closes.

See it, then try it

Watch the Advisor decide — 2 minutes, no install

Open the interactive demo and choose the Engineering Advisor scenario: a live pre-flight where the score says SHIP and the Advisor says wait — and is right. The exact moment above, played out.

Try this yourself

Connect an environment read-only and run a pre-flight check against it. On Professional and Enterprise the Advisor’s note appears beside the verdict whenever it has something material to add.
To understand what stands behind a note like that — evidence, what the system knows and doesn’t, and how to challenge it — read How ChangeGuard AI Makes Decisions.

Silence is deliberate

If the Advisor has nothing material to add beyond the score and the verdict, it stays silent. Restraint is the feature — you should never see filler advice, and when a note does appear it’s because something changed the decision.

What it raises, and what it cannot see

The Advisor reads two records the platform already holds for the target environment: its own open or analyzed incidents, and its recent change record. It raises at most the one or two things that would change the decision:
  • An active incident in the environment right now, with why it matters for this change and what would make the Advisor reconsider.
  • A notably bad recent change record on this environment, from reported outcomes.
Every point cites the incident or change record it rests on; counts are the backend’s own (a read bounded at 50 says “more than 50”, never a false exact number). The Advisor never alters the verdict, never blocks and never executes; it makes no model call. It reasons about the environment’s current state and recent record, not the specific change: it cannot see the diff, Terraform, or who is on call, and it says so in its blindspots line. It speaks only with live environment evidence (Fleet or Edge); a judgment with no environment connected reports no_environment_evidence.

Availability

The Engineering Advisor is generally available on Professional and Enterprise (plan feature advisor) and on by default there. It appears on both judgment surfaces: the dashboard check and the pre-flight call your CI makes (POST /api/cicd/preflight). An admin can turn it off for the whole workspace in Settings > Engineering assistance; every change there is audited. On the 14-day evaluation and Starter the check reports unavailable and names the plan; the verdict itself is unchanged. What every plan has, with or without the Advisor: the pre-flight check (SHIP / HOLD / BLOCK with the score and reasons) and incident investigation with cited evidence, a proposed fix and how to verify it. The Advisor is one of four engineering-intelligence capabilities, all generally available: the Engineering Opinion (an owned, evidence-tied position on an incident), Engineering Reasoning (confidence factors and why not the obvious move) and Engineering Experience (your workspace’s own history, recalled when it changes the decision). Start with the overview. Next: Understanding your score →