Backtest
Backtest
Model Score
Explains how the composite model score is calculated and how to interpret it.
The model score combines win rate and exponential-decay percentile into a single benchmark.
As of: 2026-03-08
Source artifact: backtest-metrics.snapshot.json
Generation command: python backtest.py
Score Breakdown
| Metric | Value | Formula |
|---|---|---|
| Model Score | 60.33% | 0.5 × win_rate + 0.5 × exp_decay_percentile |
| Mean Excess | +4.07% | dynamic_percentile - uniform_percentile |
| Median Excess | +4.41% | Typical outperformance |
| Relative Improvement | 11.44% | excess / uniform × 100 |
| Ratio | 1.11 | dynamic / uniform |
Performance Summary
- 64.45% win rate (1,646 wins / 908 losses)
- 11.44% relative improvement on average
- Recent weighted percentile average: 56.21%
Worked Example
Using a hypothetical win rate of 64.00% and exponential-decay percentile of 56.00%:
model_score = 0.5 * 64.00 + 0.5 * 56.00 = 60.00%
This example is interpretation guidance only. Operational caveats remain in Assumptions and Limitations.