Phase 2
Learning loop
Closes the loop. Captures every accept / reject / false-positive / missed-opportunity from human reviewers and feeds it back into the scoring weights, the suppression rules, and the threshold floors — so the model gets sharper over time without manual re-tuning.
Coming in Phase 2
What this page will include
- Reviewer feedback taxonomy: accepted · rejected · false positive · missed opportunity · wrong price · wrong product
- Per-signal contribution review — which signals were predictive in approved deals vs. rejected ones
- Suggested weight adjustments with backtest preview ("if we shift renovation +3 / distress -2, last quarter would have surfaced X more, suppressed Y more")
- Model quality dashboard: false positive rate, tier precision, surfaced-to-reviewed ratio, performance by ZIP
- Feedback audit log — every learning input with reviewer, date, outcome
Phase 1 (data foundation + training pipeline) is the focus right now. This module activates once the foundation reliably produces clean tiered flips and stable market models for every covered ZIP.