WatchlistVT-2041Active
Model quality
AnalyticsSeeded demo evaluation — not measured model performance. Potential match — requires human review.
Precision / recall vs threshold
Confusion matrix — seeded truth
A threshold experiment on seeded labels; changing it does not alter live alert rules or retrain a model.
Needs review: a group precision gap exceeds the selected limit. Small demo samples are not evidence of validated fairness.
By camera
| Group | Samples | Precision | Recall |
|---|---|---|---|
| 5 | 100.0% | 100.0% | |
| 5 | 100.0% | 60.0% | |
| 1 | — | — | |
| 4 | 100.0% | 66.7% | |
| 2 | 50.0% | 100.0% | |
| 2 | — | 0.0% |
By age group
| Group | Samples | Precision | Recall |
|---|---|---|---|
| 7 | 100.0% | 100.0% | |
| 3 | 100.0% | 100.0% | |
| 5 | 50.0% | 25.0% | |
| 4 | 100.0% | 50.0% |
By time of day
| Group | Samples | Precision | Recall |
|---|---|---|---|
| 17 | 85.7% | 66.7% | |
| 2 | 100.0% | 50.0% |
False-positive examples

AL-11 · CAM-068 · 76%
Illustrative image; seeded false-positive label.
Rejection reasons
- Wrong person0
- Wrong clothing0
- Poor image quality0
- Other0
Underlying alert records · 19
- Reviewed
- Reviewed
- Reviewed
- Rejected
- Reviewed
- Rejected
- Rejected
- Reviewed
Demo data — all mocked