Crate kcode_audio_speaker_review

Expand description

Human speaker-review policy and classifier application for audio packets.

Structs§

CandidateMapping: Classifier evidence retained for one chunk-local speaker.
ChunkConfirmation: Exact speaker resolutions and signoff for one review chunk.
CorrectionChunk: One complete chunk in a recording-level correction packet.
CorrectionObservation: One deterministic classifier observation in a correction packet.
CorrectionPacket: Complete transport-neutral correction packet for one recording.
ObservationConfirmation: One observation-level human resolution.
ObservationKey: One caller-owned observation identity.
ParsedChunk: Complete normalized speaker structure for one raw provider result.
ParsedSpeaker: One typed speaker row normalized from a raw provider response.

ConfirmationState: Durable identity-confirmation lifecycle for a correction packet.
LegacyReviewDisposition: Durable resolution for a finalized recording with an obsolete unsigned packet.
SpeakerResolution: Human resolution for one chunk-local speaker.

CLASSIFIER_MODEL: Exact classifier model cohort component.
CLASSIFIER_PROMPT_VERSION: Exact classifier prompt-version cohort component.
CLASSIFIER_PROVIDER: Exact classifier provider cohort component.
CLASSIFIER_SCHEMA_VERSION: Exact classifier feature-schema cohort component.

apply_confirmation: Applies one exact chunk signoff to the classifier and in-memory packet.
confirmation_matches: Reports whether a signed chunk already has the exact submitted mapping.
legacy_review_disposition: Chooses the durable resolution for an obsolete finalized review packet.
restore_training: Best-effort restores classifier rows to one packet’s authoritative state.
training_object_id: Returns the deterministic training object identity for one review chunk.
validate_confirmation_coverage: Validates exact, complete observation coverage for one chunk signoff.