Rule-based safeguards ship free and always run. Pro adds scoring on top of them, never instead of them.
The scorer decides
Fraud decisions come from a deterministic PHP scorer with transparent weights you can be shown — click velocity, self-referral proximity, coupon abuse patterns, IP and device repetition. A model may put a verdict into words for you, but no model sits on the path that gates a commission.
It fails open
If scoring is unavailable, misconfigured, or returns something the plugin does not recognise, the referral is approved and marked degraded. An affiliate is never left unpaid because a scoring service had a bad day.
This is deliberate and it is not configurable. The alternative — failing closed — means a provider outage silently stops crediting every affiliate you have, and you find out from a complaint rather than from a log.
What a score is for
A score sorts your review queue. It marks a referral as worth a look and tells you which signals fired. It does not reject anything, and the rejection is always yours.
Your own AI provider
Where Pro generates text it goes through WordPress core’s AI Client, so you pick the provider by installing a connector and supplying your own key. There is no bundled model SDK, no service of ours in the path, and no API key in our code. If you install no connector, the generative features are simply absent and everything else works.