Machine Learning & Hybrid Strategy Studio
Immutable model registry, point-in-time feature explainability, shadow inference scoring, and continuous PSI drift monitoring.
Immutable Model Registry & Lineage
Strict 6-stage lifecycle: EXPERIMENT → CANDIDATE → SHADOW → PAPER → LIVE → RETIRED
EUR/USD ICT Setup Classifier
EURUSDStatistical classifier predicting ICT setup win-rate using point-in-time order flow and FVG features.
| Version | Stage | Validation ROC-AUC | Accuracy / F1 | Dataset SHA-256 | Actions |
|---|---|---|---|---|---|
| v1.0.0 | Shadow Mode | 69.8%(Brier: 0.215) | 64.2% / 63.6% | e3b0c44298fc... |
Shadow Model Prediction & Explainability
EURUSDReal-time inference attribution without direct order execution authority
Latency: 1.25msHIGH Confidence
Model Setup Probability
72.5%Bullish Setup
Decision: Class 1
Time: 8:15:00 AM
Top Contributing Feature Weights (Log-Odds Attribution)
fvg_gap_size+0.420 (BULLISH)
Raw input value: 0.0018
return_5+0.280 (BULLISH)
Raw input value: 0.0012
is_london_killzone+0.180 (BULLISH)
Raw input value: 1.0000
candle_range_norm-0.090 (BEARISH)
Raw input value: 0.0025
Covariate & Concept Drift Monitor
Distribution StablePopulation Stability Index (PSI) tracking distribution shifts between training baseline and live inference
Checked: 8:15:00 AM
Overall PSI
0.0420
< 0.10 (Nominal)
Baseline Samples
500
Training dataset
Inference Window
100
Live sliding window
Recommended Action
MAINTAIN
Individual Feature Drift Metrics
| Feature | PSI Score | Status |
|---|---|---|
| return_1 | 0.0310 | STABLE |
| return_5 | 0.0480 | STABLE |
| fvg_gap_size | 0.0520 | STABLE |
| candle_range_norm | 0.0380 | STABLE |