tier_verdicts. The fields below support logging, reviewer interfaces, and certification records.
Scores
| Field | Description |
|---|---|
truth_score | Overall AI probability. 0 = human, 1 = AI. Use for displaying confidence to reviewers |
geo_score | Secondary corroborating score, stable across audio processing variations |
proto_score_v1 | Drives the press-safe tier. Include in certification records |
geometry_fused_score | Internal score. Available for logging |
head_a – head_k | Individual detection signal outputs. Each is 0–1, higher = more AI evidence. Use for logging or reviewer interfaces |
embed_temporal_std | Acoustic consistency across the track. Available for logging |
reliability | Confidence in the result based on audio quality and duration. Below 0.6 — treat with extra caution |
Origin attribution
| Field | Description |
|---|---|
origin | Predicted generator: "suno", "udio", "elevenlabs", "human", "unknown" |
origin_confidence | Confidence in the prediction (0–1) |
verdict or tier_verdicts.
Audit fields
| Field | Description |
|---|---|
calibration_version | Version of the model calibration that produced this result. Include in certification records |
processed_at | ISO timestamp of when analysis completed |