Tested for disparate impact, on a schedule.
A synthetic bias probe varies demographic signals while holding qualifications constant, then measures selection rates across groups. It runs before and after deployment on a schedule, re-testing whenever the scoring configuration changes and at least once a year regardless. An adverse finding pauses Amp acting on its own scores for that job until an admin reviews it.
Selection rates, intersectionally. One cell fails, everything pauses.
Synthetic candidates: identical qualifications, varied demographic signals, run through the same production gate as real applicants. If any group is selected at less than 80% of the top group's rate, and the gap is statistically real, the run reads adverse. That cell shows at a glance, and the job goes on hold.
On a schedule, not on request.
Testing runs whether or not anyone asks: a baseline before deployment, on a schedule after that, immediately on any change to the scoring configuration, and a fresh-run floor of once a year. Admins set how much of the job list each scheduled run covers.
See Amp running.
A 30-minute walkthrough, keyed to one of your roles and one of your geographies.
See Amp running