Model Accuracy
0.0%
β stable this week
Source: demo baseline
Dataset split: 10,455 train / 485 val / 2,289 test (~13,229 total, 24 PII categories, 6 data sources)
Macro AUPRC
Mean PR-AUC (low / med / high). This is not ROC-AUC.
0.000
Good
β current report snapshot
Source: demo baseline
False positive rate (holdout)
0.0%
Safeβrisky rate
Label FPR at Ο (not the ~4% miss rate)
Source: demo events
Precision
0.000
Operating point
β robust on validation sweep
Source: demo baseline
Recall (risky) / TPR
0.000
Catch rate for risky class
Holdout @ recommended threshold
Source: model_records.json
False negative rate (risky)
0.0%
Missed risky rate
FNR = 1 β recall
Not the same as FPR (safe text flagged as risky).
Calibration FPR (validation)
0.0%
Threshold sweep on val
Lower than holdout FPR
From threshold_calibration.json β not paste telemetry.
Precision-Recall Curve
Live PR data unavailable
PR curve
Eval OP (red)
β Holdout Ο (green)
Min precision target
ROC (operating point)
ROC-AUC: N/A β ROC-AUC not in summary yet. Red point = calibrated FPR vs recall (high). Headline KPI is macro PR-AUC, not ROC-AUC.
ROC
Random classifier
Eval OP (red)
β Holdout Ο (green)
Confusion Matrix (Predicted vs Actual)
Reference matrix shown (live confusion export unavailable).
Performance by PII Class
Synthetic reference rows (not live production telemetry).
| PII Class | Sensitivity | Expected | Actual | Score Range | Status |
|---|
Score Distribution Heatmap (Current vs Previous Week)
Drift Detection (Weekly Correction Rate)
Correction rate %
Warning threshold: 8%
Retrain trigger: 15%
Threshold Simulator
Adjust warn/block cutoffs and preview policy impact.
20-60
50-95
Simulated block: 409 (+123 vs current)
Simulated warn: 34 (-132 vs current)
Simulated safe: 557 (+9 vs current)
Estimated FPR proxy: ~0.0%
Copies suggested values for risk_policy.json.
Live FPR Monitor
Displayed metric: FPR proxy
FPR value: 0.0%
Holdout FPR: N/A
Calibration FPR: N/A
Total blocks: 286
Total snoozed: 0
Override rate: 0.0%
Target < 5%, alert > 10%
Active Learning Status
No feedback rows synced yet. Collect corrections from remediation dialog to populate active learning stats.
False Positive Log (Recent 20)
| Time | Score | PII Class | Action | Feedback |
|---|