Don't take our word for it
— read the measurements.

We run a fixture lab of hand-built, ground-truth-labeled scenarios against our own detection logic on every change, and publish the result. A committed floor value fails the build if a module's measured precision or recall ever regresses.

Every scan we run publishes a coverage manifest — "X of Y checks ran" — instead of letting a skipped check silently read as a pass. A check can land in one of a few honest states:

Checked, clean

The check ran and found nothing to flag.

Checked, found (and fixed or flagged)

The check ran, found a real issue, and either fixed it automatically or handed you the exact next step.

Checked, evidence attached

Where we can, we keep proof — a captured response, a screenshot — instead of asking you to trust a green result blindly.

Not checked

A data source isn't connected, the check needs a higher plan, or it's temporarily unavailable. We show this out loud instead of letting a skipped check read as "passed".

Not applicable

The check doesn't apply to your stack — e.g. a WordPress-specific check on a site that isn't running WordPress.

Inconclusive

We probed, but the signal wasn't clear enough to call it either way — flagged as inconclusive rather than forced into a false "clean" or a false alarm.

We show "not checked" out loud because a false sense of coverage is more dangerous than an honest gap — you can act on a gap you know about.