Identify the constraint: Policy, practice, or pipeline

A number moving in the wrong direction usually comes from one of these, and each takes a different fix:

• A miscalibrated policy. The rule fires on work your organization considers acceptable.

• A practice problem. Teams author changes that genuinely violate standards.

• Pipeline mechanics. Checks run in the wrong place, or too late to be cheap.

They look the same on a tile, because a tile totals everything in scope. Naming which one you are looking at takes a few specific readings, and it is worth doing before you change anything.

Read the distribution, not the total

The Policy Checks dashboard shows which checks fired, at what severity, and against which databases and changelogs. How those findings are spread is the diagnosis.

• The same check failing in every project points at the policy. If everyone violates a rule, the rule is describing work your organization actually does, or its threshold is wrong. Filter the dashboard to one project at a time to see this, since nothing groups findings by team. See Compare applications, teams, and environments.

• Many different checks failing in one project points at practice. The rules hold everywhere else, so the fix lives in how that team authors and reviews changes. Filter to that project and scan the Severity column down the operations list to see how wide the spread is.

• Findings that only ever appear on deployment runs point at the pipeline. Filter the operations list by Check and look at what else ran. If the only operations that catch a rule are the ones triggered by a deployment, enforcement is happening at the most expensive moment rather than in the developer loop.

• Passing checks alongside drift or failed deployments points at coverage. Nothing fired because nothing was watching. Read Database Coverage and Changelog Coverage beside the pass rate, and check the Drift Detection dashboard for databases nothing is checking. See Review policy coverage.

Note: The product does not track production incidents, so a rule passing while something still went wrong is not something a dashboard will tell you. Drift appearing on a database whose checks all pass is the closest signal you can read directly.