Trust
What we score, and what we never score.
Published here, in the open, and enforced directly in the real scoring code.
Excluded from scoring
Appearance, dress, background
Accent, pronunciation, grammar
Name, school, employer prestige
Employment gaps
Body language, eye contact, posture
Can I trust an AI’s read on a candidate?
The Dril Card is evidence for you to review and decide on. Every score ships with the exact quote it’s based on, so you’re reading the tape yourself.
“The RAG eval pipeline flagged silent regressions before they ever hit production. That’s the whole reason I built it that way.”
Five rules, enforced in the product.
Every score is transcript-backed
Unquotable means unscored. A dimension with no evidence renders as Unscored.
A human decides
A Dril Card is a recommendation. A human reviews every shortlist and makes the call.
Consent, per posting
A candidate agrees to be driled for each specific role, every time.
Rubrics are versioned
When scoring changes, old Dril Cards keep the rubric they were scored against.
No system is unbeatable
We flag tab-switches and leaving fullscreen, the signals we actually catch.
Before you decide.
What hiring teams usually ask about trusting this.
Yes. Consent is per posting and explicit, every time. A candidate agrees to be driled for a specific role before that Dril's evidence ever reaches you. There's no blanket opt-in that covers every posting; each one asks again, so a candidate always knows exactly which role their evidence is tied to.
We flag tab-switches and leaving fullscreen. A flag doesn't auto-reject anyone. You decide what it means alongside the rest of the evidence.
No. Each Dril Card keeps the rubric it was scored against, so a rubric change never reaches back and changes a decision on someone already evaluated.
Yes. Every dimension either shows the exact quote it's based on, or renders as Unscored if there wasn't one to cite.