fix: comparison report winner logic — use common metrics only (#120)

openagent_eval/reports/comparison.py

@@ -99,28 +99,42 @@ def generate(self, report: Any) -> str:
             lines.append(f"  Metrics unchanged: {len(all_metrics) - improved - regressed}")
             lines.append("")
 
-            # Overall scores
-            if baseline_metrics:
-                base_overall = sum(baseline_metrics.values()) / len(baseline_metrics)
-                lines.append(f"  Baseline overall:    {base_overall:.4f}")
-            if experiment_metrics:
-                exp_overall = sum(experiment_metrics.values()) / len(experiment_metrics)
-                lines.append(f"  Experiment overall:  {exp_overall:.4f}")
-                if baseline_metrics:
+            # Overall scores (common metrics only for fair comparison)
+            if baseline_metrics and experiment_metrics:
+                common_metrics = sorted(
+                    set(baseline_metrics) & set(experiment_metrics)
+                )
+                if common_metrics:
+                    base_overall = (
+                        sum(baseline_metrics[m] for m in common_metrics)
+                        / len(common_metrics)
+                    )
+                    exp_overall = (
+                        sum(experiment_metrics[m] for m in common_metrics)
+                        / len(common_metrics)
+                    )
+                    lines.append(f"  Baseline overall:    {base_overall:.4f}")
+                    lines.append(f"  Experiment overall:  {exp_overall:.4f}")
                     overall_delta = exp_overall - base_overall
                     lines.append(f"  Overall delta:       {overall_delta:+.4f}")
-            lines.append("")
-
-            # Winner
-            if baseline_metrics and experiment_metrics:
-                base_overall = sum(baseline_metrics.values()) / len(baseline_metrics)
-                exp_overall = sum(experiment_metrics.values()) / len(experiment_metrics)
-                if exp_overall > base_overall:
-                    lines.append(f"  >> WINNER: {report.experiment_name}")
-                elif exp_overall < base_overall:
-                    lines.append(f"  >> WINNER: {report.baseline_name}")
+                    lines.append("")
+
+                    # Winner
+                    if exp_overall > base_overall:
+                        lines.append(f"  >> WINNER: {report.experiment_name}")
+                    elif exp_overall < base_overall:
+                        lines.append(f"  >> WINNER: {report.baseline_name}")
+                    else:
+                        lines.append("  >> TIE")
                 else:
-                    lines.append("  >> TIE")
+                    lines.append("  No common metrics to compare overall scores.")
+                    lines.append("")
+            elif baseline_metrics:
+                lines.append(f"  Baseline overall:    {sum(baseline_metrics.values()) / len(baseline_metrics):.4f}")
+                lines.append("")
+            elif experiment_metrics:
+                lines.append(f"  Experiment overall:  {sum(experiment_metrics.values()) / len(experiment_metrics):.4f}")
+                lines.append("")
         else:
             lines.append("  No metrics available for comparison.")
             lines.append("")