Documentation
FaceCore V1 Documentation
FaceCore is a biometric verification engine focused on calibrated matching, quality gating, and presentation attack detection. It stores zero government ID records and leaves billing and identity policy to your application.
Deterministic Decision States
Every stage in the pipeline fails safely. If image quality is below q_min = 40.0, or a model is unavailable, FaceCore never guesses or defaults to MATCH.
Quality passed, liveness passed (when requested), and similarity >= calibrated threshold.
Quality and liveness passed, but similarity fell below the calibrated threshold.
Frame failed quality gating (e.g. BLURRY, TOO_DARK) or liveness fell in the uncertain band. Safe to retry.
Presentation attack detected (SPOOF_SUSPECTED / PAD_FAILED) or multi-view 3D check flagged a tilted photo.
Calibration Profiles
Instead of guessing magic cosine similarity numbers, choose a profile calibrated against your embedder and capture type (selfie_selfie, selfie_id, or multiview).
| Profile | Target FAR | Example Threshold | Recommended For |
|---|---|---|---|
| strict | 0.01% FAR (1 in 10,000) | 0.562 | High-value account recovery, privileged access, financial step-up. |
| standard | 0.1% FAR (1 in 1,000) | 0.452 | Default onboarding, selfie-to-ID checks, workforce sign-in. |
| relaxed | 1.0% FAR (1 in 100) | 0.348 | Low-friction convenience re-authentication paired with active liveness. |