Use expected_output (reference[0]) as Task description instead of truncated prompt

app.py

@@ -43,8 +43,9 @@ def load_results() -> pd.DataFrame:
                     usage = rec.get("token_usage") or {}
                     ev = rec.get("evaluation") or {}
 
-                    raw_prompt = (rec.get("input") or {}).get("raw", "")
-                    task = (raw_prompt[:97] + "…") if len(raw_prompt) > 100 else raw_prompt
+                    inp = rec.get("input") or {}
+                    reference = inp.get("reference") or []
+                    task = reference[0] if reference and reference[0] else inp.get("raw", "")[:100]
 
                     rows.append({
                         "Namespace":      meta.get("eval_namespace", "?"),