Get an interaction (the evidence view)
The full per-call scorecard (findings with plain-English reasons + cited regulations, satisfied requirements, redactions, gate health, verdict) plus the tamper-evident audit_hash. With scorecard-v1 the response also carries the optional per-call timeline and quality_metrics eval blocks (empty until the engine emits them) and derived_metrics — the platform’s score-ready rollup of the timeline (TTFB, turn counts, barge detect + recovery). Requires the trace:read scope.
Authorizations
Use Authorization: Bearer pyai_live_... (or pyai_test_...).
Path Parameters
The call_id.
Response
Interaction detail
"trace.interaction"
call_id
hear, voice, speak, clone, cue, flow, omni, agents, null PASS, WARN, FAIL Unix ms.
Hash-chain link proving this record is unaltered (the signed-evidence guarantee).
The full trace-scorecard-v0/v1 record: findings, satisfied requirements, redactions, gate health, verdict, and (scorecard-v1) the optional timeline/quality_metrics eval blocks.
Tier-2 (async semantic) findings for this call, joined at read time. They arrive after the scorecard and never alter audit_hash.
scorecard-v1 per-call timeline (transcript turns + latency stamps used for eval scoring), hoisted from the scorecard for convenience. Empty [] until the engine emits it.
scorecard-v1 per-call rolled-up quality metrics, hoisted from the scorecard. null until the engine emits it.
Platform-computed rollup of the timeline into score-ready eval aggregates (TTFB, turn counts, barge detect + recovery). Always present; zero counts and null percentiles until the engine emits a timeline. Percentiles use the same method as the offline eval harness, so an online per-call score lines up with the offline benchmark.