fix: handle missing metric values in EvalResult by replacing None with NaN

src/leaderboard/read_evals.py

@@ -109,7 +109,10 @@ class EvalResult(BaseModel):
             # TODO: support multiple metrics
             metric_keys = ["caa"]
             accs = np.array([
-                v.get(metric_key, None) for k, v in data.results.items() if task.key == k for metric_key in metric_keys
+                v.get(metric_key, np.nan)
+                for k, v in data.results.items()
+                if task.key == k
+                for metric_key in metric_keys
             ])
             if accs.size == 0 or any(acc is None for acc in accs):
                 continue