A forensic pipeline, not a black box

Every scan runs the same six-stage pipeline; every stage's output is inspectable and downloadable.

1. STEP 01
   Ingest

   Video uploaded via console or API is stored in a private, org-scoped bucket. Only signed URLs, short TTL.

2. STEP 02
   Sample

   Frames are extracted at deterministic timestamps. Audio waveform sampled in parallel.

3. STEP 03
   Analyze

   Six detectors run in parallel: frame artifacts, temporal consistency, lip-sync, audio synthesis, C2PA/provenance, compression forensics.

4. STEP 04
   Aggregate

   Each signal returns a 0-1 score, calibrated confidence, and structured evidence. A weighted aggregator produces the final verdict.

5. STEP 05
   Review

   High-risk verdicts route into your analyst queue. Overrides are audited.

6. STEP 06
   Return

   The verdict, per-signal breakdown, frame evidence, and confidence band are available via UI, JSON API, and webhook.