Inspecting results

Four pages show what the platform stored for your tenant, in the selected mode, so a reviewer can open a single check without going through the API. What appears depends on your storage policy: results need store_results, document images need store_source_images, and liveness frames need store_liveness_frames.

Liveness checks

/liveness-checks lists every POST /v1/liveness/check with its time, mode (passive, burst, challenge), verdict, score against your threshold, frame count, reason and customer_ref. Opening a row shows:

• the captured frames, when frames are stored, in the order they were submitted, with the frames the anti-spoof models scored marked;
• the verdict: the reason with a one-line explanation, the score against your threshold, the model version and, for a challenge, the action that was issued;
• when component data was stored (always in test mode, in live mode when liveness_show_components is on for your tenant): each model's score against its threshold, the continuity check across frames (identical frames, identity, landmark jumps), the challenge action with what the verifier measured on every frame and which frame served as the baseline, and the face size, sharpness and brightness of the first frame. The raw block is one click away.

This is the page to open first when an integrator reports that genuine users are failing: the frames show whether the capture was upright, whether the last frame is a full profile, and whether the action was performed in the direction the server expects. See Liveness checks for what each reason means.

OCR scans

/ocr-scans lists every POST /v1/ocr with its document type, detection confidence, whether images are stored and customer_ref. A scan's page shows the captured images and, for a 2026 CID or SRC, the photograph decoded from the back QR code; the parsed fields as a table, or for a refused scan the fields the extraction read before the gates refused it; the eligibility verdict with every gate rejection; the raw OCR text; and the decoded back barcode with its cross-check against the printed text.

Fraud checks

/fraud-checks lists every POST /v1/fraud/detect with its decision, image heuristic score, document type and customer_ref, filterable by decision. A check's page is the review page: the decision with the reasons it rested on, the image heuristic score (diagnostic; at or above your review threshold it sends a check to review), the document images with the regions the checks reported drawn on them (red for checks that can change the decision, dotted grey for supporting measurements), the four image signals with each check's status and score, the field validation over the OCR result, the deterministic document evidence, the image quality gate per side, and earlier checks that share the same document number, image or portrait. Decisions are explained on Fraud detection.

Face registrations

/face-registrations lists the gallery entries created by POST /v1/face/registrations: when each was registered, its customer_ref, the model version of its embedding and whether the enrolled selfie is stored. A registration's page shows that selfie with the detected bounding box and quality metrics. Registrations from an older model version cannot be compared with the current one (model_version_mismatch on the API) and need re-registering.

All four pages follow the live ⇄ test toggle and expire rows according to the TTLs in your storage policy.