Model Tuning & Calibration
Tunes detection thresholds and backtests surveillance models to reduce false positives while preserving coverage
Models Managed
11
Blended False Positive Rate
6.2%
Calibrations Due
2
Backtests Run (30d)
24
| Model | False Positive Rate | Precision | Last Calibrated | Status |
|---|---|---|---|---|
| Spoofing & Layering Detection | 6.8% | 91.4% | Sep 20, 2026 | Current |
| Insider Trading Detection | 4.1% | 95.2% | Sep 15, 2026 | Current |
| Wash Trading & Marking the Close | 9.2% | 87.6% | Aug 28, 2026 | Calibration Due |
| Cross-Market Surveillance | 11.4% | 82.1% | Aug 12, 2026 | Calibration Due |
| Communications Surveillance | 14.8% | 78.9% | Jul 30, 2026 | Recalibrating |
About This Agent
Data on This Page
Every detection model's false positive rate, precision and last calibration date, with a current/due/recalibrating status. Stat tiles cover models managed (11), the blended false positive rate across all models (6.2%), calibrations currently due, and backtests run in the last 30 days (24).
What This Agent Does
Backtests every detection model's thresholds against labelled historical alert outcomes (confirmed abuse vs. closed-no-action), recommends threshold adjustments that improve precision without materially reducing recall on confirmed patterns, and schedules periodic recalibration so models stay tuned as market microstructure and trading behaviour evolve.
Worked Examples
- Spoofing & Layering Detection's false positive rate fell from 8.9% to 6.8% after this month's calibration tightened the cancel-rate threshold for high-frequency market-making accounts specifically.
- Communications Surveillance is mid-recalibration after its false positive rate crept to 14.8% — the lexicon is being retrained on six months of analyst-labelled closed alerts to cut noise from benign trading-desk banter.
- Cross-Market Surveillance is flagged for calibration due to its comparatively low 82.1% precision, the lowest of the 11 managed models, next in the recalibration queue.