# SightRadar Glossary

> Definitions of face-recognition and API terms used across SightRadar's docs.

Source: https://sightradar.com/glossary

- **Faceprint / embedding** — a numeric vector derived from a detected face, used to measure similarity between faces.
- **Collection** — a per-tenant namespace that stored faces live in (e.g. one per event). Creating one is free.
- **Index** — detect quality-gated faces in a photo and store their embeddings in a collection.
- **Search** — find every photo a person appears in from a single selfie, ranked by similarity.
- **Compare** — 1:1 similarity between two faces for identity verification.
- **Detect** — locate and quality-gate faces without storing anything.
- **Similarity score** — a calibrated 0–100 score you threshold to decide a match.
- **Similarity threshold** — the cutoff you choose; matches above it are returned.
- **Match tier (auto / review)** — every search match is labelled `auto` (high-confidence, safe to reveal) or `review` (plausible, show behind a confirm step). Both are real matches; the tier only guides how a client presents them.
- **Quality gate** — a minimum bar for face size, sharpness, pose, and detection confidence; faces that fail are rejected rather than stored as low-quality faceprints.
- **False-match rate (FMR)** — the rate at which different people are wrongly scored as the same; you tune it by choosing your threshold.
- **Idempotency key** — a header that makes a billable call safe to retry; a replay under the same key returns 409 before any charge.
- **Per-tenant isolation** — one account's face vectors are never cross-searchable with another's.

Full glossary: https://sightradar.com/glossary