Uncertain forecasts

Does the loop actually improve the model? Honest answer: not yet.

Loop state: plumbed_idle. The loop is wired end to end (queue -> submit-label -> aggregate -> promote -> retrain) but 0 human labels have been submitted. Until reviewers label candidates at /atlas/active-learning the loop is idle by design — value depends on humans actually labelling. We surface the simulated lift even when it is negative.

Queue: 405 candidates · Human labels submitted: 0

• Active-learning value depends on humans actually labelling. If labels_submitted is 0 the loop is plumbed but idle.
• simulated_f1_lift uses PROXY labels (incident-table positives within +/-3 days, else negative), not human labels. It is an optimistic estimate — real human labels on uncertain days are noisier and the AL queue surfaces days the incident table is silent on. Treat the lift as 'is the queue surfacing learnable rows', not a promotion signal.

Live source: /v1/sentinel/active-learning-loop-status