OMRE Open Metrics
We don't hide our math. We display it.
A real-time audit of the OMRE Stability Engine, updated every 24 hours.
AI Engine Status
Checking...
Data Points Analyzed
0
12 Years of History × Nifty 500
Search Latency
0ms
Vector Search Response Time
Risk Filter
0 Stocks Blocked
Rejected due to low liquidity/high debt
Feature Dictionary
The OMRE 5D/15D/25D models use 26 features: 20 Base V2 + 6 Swing (for 5D accuracy).
Base V2 features (20)
- rsi_14 — Standard momentum; overbought/oversold
- ema_50_div — Distance from 50-EMA; trend extension
- ema_200_div — Distance from 200-EMA; core trend health
- atr_14_norm — Normalized volatility; adapts targets
- rvol — Relative volume; institutional interest
- log_return — Raw price momentum (24h)
- rel_strength — Performance vs Nifty 50 (Alpha core)
- adx_14 — Trend strength; filters choppy markets
- bb_width — Volatility squeeze; breakout precursor
- dist_52wh — Proximity to 52-week high
- momentum_strength — Short-term velocity
- panic_buy_signal — Volume spurt + price jump
- ema_50_zscore — Extension vs 50-day normal
- trend_regime — Global bull/bear flag
- is_breakout — Range breakout detection
- pattern_doji, pattern_engulfing, pattern_hammer, pattern_morning_star, pattern_shooting_star — Candlestick patterns for entry timing
Swing features (6) — 5D boost
- rsi_12 — Faster momentum (optimal for 5D)
- ema10_div — Short-term pullback entry
- ema10_zscore — Price extension vs 10-day volatility
- roc_10 — Rate of change (2 weeks)
- macd — Trend convergence/divergence
- macdsignal — Momentum crossover signal
5D Swing Accuracy (V2 Production)
Walk-forward: Train 2014–2024, Test Jan 2025+. Top 5 picks/day, target +3% in 5 days. Filter: RSI 30–80, RVOL > 0.5, RS/Mom > 0. Source: accuracy_metrics.
2025 Full
50.8%
Jan 2026
46.7%
Feb 2026 (MTD)
73.3%
Precision Matrix
Tiered accuracy across all prediction horizons (from accuracy_metrics). Intraday shows Top 1–15.
| Tier | Hit Rate | Avg Return | Correct / Total | Days |
|---|---|---|---|---|
| No data available for this horizon yet. | ||||
Horizon Comparison
What Counts as a "Hit"?
Backtest Methodology
- Train period: 2014–2024 (10 years). Model learns structural breakout patterns and volatility regimes.
- Test period: Jan 2025–Feb 2026 (walk-forward, out-of-sample). Metrics reflect real-world predictive power.
- Classification: Balanced XGBoost. Stocks ranked by "Success Probability"; Top 1 = highest conviction for that day.
Adaptive Best-Fit Portfolio (Buy-Only Filter)
The engine shows only picks with a BUY signal. Neutral/Avoid are hidden. Budget is allocated by optimal N slots (e.g. Top 4); if a slot is skipped (Neutral), that capital stays as CASH and is not reallocated. Data source: best_fit_portfolio.
Database Table Locations (Cloud)
OMRE tables used for predictions, accuracy, and portfolio. Synced daily to cloud.
| Data category | Table name | Use |
|---|---|---|
| Main predictions (5D, 15D, 25D) | predictions | Swing scores |
| 1D Alpha Engine (full score) | predictions_1d | Intraday raw |
| 1D XGBoost picks (Top 15) | predictions_xgb_picks | Intraday use this |
| Accuracy / precision | accuracy_metrics | Accuracy use this |
| Portfolio allocation (buy-only) | best_fit_portfolio | Best-fit use this |
| Technical features (OHLCV + 35) | stock_history | Features |
| News & sentiment | stock_news | Sentiment |
Live Decision Log
Usage of this data is subject to OMRE's terms of service.