Hybrid Human In The Loop Review Qa

An ED physician finishes a chest-pain workup encounter. The ADT feed carries patient_class=E, location=ED_BAY_4, chief_complaint=chest_pain. The audio classifier returns encounter_type=ED with confidence 0.94 based on phrases lik…

Agent emits a single CLASSIFICATION_COMPLETE event with queue_destination=ED_REVIEW_POOL, template_id set to an ED-designated template, sla_tier=ED_4HR, and sla_start_time equal to ingestion_timestamp. Classification evidence record stores both audio_signal=ED and adt_signal=ED. No human escalation…

A patient is boarded in an ED hallway for chronic low back pain — an ambulatory complaint. ADT feed shows patient_class=E, location=ED_HALLWAY, chief_complaint=LOW_BACK_PAIN. Dictation language is 'patient returns today for follo…

Agent detects signal_conflict=true and does NOT emit CLASSIFICATION_COMPLETE. It emits CLASSIFICATION_CONFLICT preserving both classifier outputs verbatim. Encounter status is set to PENDING_ADJUDICATION. SLA clock does not start. A human adjudicator queue entry is created containing both classifie…

A patient has AI_TRANSCRIPTION_OPTOUT=true in their EHR consent record. Their clinician submits a recording. The ingestion pipeline must read consent flags as the first processing step, before any call to the audio ML classifier …

Agent reads consent_flags as the first processing step, before any classifier invocation. Sets processing_mode=HUMAN_ONLY and emits CONSENT_BLOCK event. Zero calls are made to the audio ML classifier API. Audio file reference is forwarded to HUMAN_ONLY_DOCUMENTATION queue. Audit log records: INGEST…