Dataset readiness · BACKTEST · Run #1
Official production ML history: BACKTEST Run #1.
Other BACKTEST runs remain stored as experiments and are not pooled into these readiness figures.
4,662Historical signals
415Stocks represented
1,556Rows usable for first success model
1,556Rows usable for false-breakout model
1,561Rows with 20-session return
100.0%Historical RS coverage
Strong starting dataset
The first XGBoost classifier uses only resolved triggered outcomes:
TARGET = 1, STOP/TIMEOUT = 0. PENDING, AMBIGUOUS and NOT_TRIGGERED are excluded.
NIFTY 500 historical breadth table detected. Breadth columns are exported with a
membership-quality flag. Approximate historical membership is not silently treated as pristine data.
Time coverage
2024-10-09 → 2026-09-25.
Training will be split chronologically, never by random row shuffle.
2024 · 876
2025 · 2,062
2026 · 1,724
What happens next
1. Complete Phase 1B historical RS / breadth enrichment
2. Export ML Dataset v1B CSV
3. Train Logistic Regression baseline
4. Train XGBoost classifier on the same chronological split
5. Compare ROC-AUC, PR-AUC, Brier score, precision and top-decile performance
6. Save model + feature importance
7. Only if XGBoost beats the baseline → integrate predictions into OmmAlpha