> ## Documentation Index
> Fetch the complete documentation index at: https://docs.worthcall.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Evaluation Tools

> 3-model ensemble scoring, outcomes, drift detection, and weight management.

# Evaluation Tools

## Evaluation data

| Tool             | Description                                                    |
| ---------------- | -------------------------------------------------------------- |
| `eval_stats`     | Total evaluations, average scores, win rate, model accuracy    |
| `eval_outcomes`  | Evaluations joined with trade outcomes (scores + R-multiples)  |
| `eval_reasoning` | Per-model key drivers, risk factors, uncertainties, conviction |
| `record_outcome` | Record trade result for an evaluation                          |

## Drift detection

| Tool           | Description                                                           |
| -------------- | --------------------------------------------------------------------- |
| `drift_report` | Rolling accuracy, calibration error by score decile, regime detection |
| `drift_alerts` | Recent alerts when accuracy fell below thresholds                     |
| `drift_check`  | Run drift report + alert check in one call                            |

## Weight management

| Tool               | Description                                              |
| ------------------ | -------------------------------------------------------- |
| `simulate_weights` | Test different model weights against historical data     |
| `weight_history`   | Audit trail of weight changes                            |
| `tune_risk_params` | Auto-tune using half-Kelly sizing from last 100 outcomes |

## Edge validation

| Tool           | Description                                                                         |
| -------------- | ----------------------------------------------------------------------------------- |
| `edge_report`  | Sharpe, Sortino, win rate, profit factor, max DD, feature attribution, walk-forward |
| `walk_forward` | Walk-forward backtest with train/test windows                                       |