Human pending
0
Opening page…
Human-in-the-loop AI governance
Review every voice transcript, export governed datasets, evaluate offline candidates, and require human approval before promotion.
Every model transcript enters HUMAN_PENDING. There is no confidence or accuracy auto-accept path. The model transcript is immutable; corrections are stored separately.
Human pending
0
Validated pending
0
Dataset versions
0
Candidate approvals
0
Auto deploy
OFF
Training is offline. Promotion requires fixed-test non-regression and a separate human.
Frozen test: created on first eligible export · Active registry model: not promoted
Listen, correct, save as HUMAN_REVIEWED, then explicitly validate or reject.
No reviews in this state.
Use this audited backfill path after an uploaded audio artifact has been transcribed. The sample still starts at HUMAN_PENDING and cannot bypass review.
Export only VALIDATED + PENDING + audio-present + reviewer-present + deduplicated samples.
Export waits until the configured validated-sample threshold is met.
No dataset export yet.
The control plane records the job. Execute training in the approved offline environment.