scripts: add scripts/compute_bestofn_with_selector.py

scripts/compute_bestofn_with_selector.py

@@ -118,6 +118,9 @@ def main():
     parser.add_argument("rollout_jsonl")
     parser.add_argument("label")
     parser.add_argument("--selector_host", default="http://localhost:8103")
+    parser.add_argument("--row_preview", action="store_true",
+                        help="Use 'OK. Rows preview: ...' format (matches v2 selector training data). "
+                             "Default uses bare 'OK' / '<error>' which matches v1 selector.")
     args = parser.parse_args()
 
     n_q = 0
@@ -188,19 +191,21 @@ def main():
                 n_majority += 1
 
         # Trained selector: score each candidate using REAL execution result (no gold-label leak).
-        # Re-use the exec_results computed above for rule-based majority.
+        # Prompt format MUST match the selector training data. Two options:
+        #   --row_preview off (default): bare "OK" / "<error>" — matches v1 selector
+        #   --row_preview on:            "OK. Rows preview: <rows>" / "Error: <msg>" — matches v2 selector
         scores = []
         with ThreadPoolExecutor(max_workers=8) as exe:
             def _make_prompt(idx, t_):
                 er = exec_results[idx]
-                if er[1]:
-                    exec_str = f"Error: {er[0]}"
-                else:
-                    rows_str = str(er[0])
-                    if not rows_str.strip() or rows_str.strip() == "[]":
-                        exec_str = "OK. Result rows (preview): (no rows)"
+                if args.row_preview:
+                    if er[1]:
+                        exec_str = f"Error: {str(er[0])[:180]}"
                     else:
-                        exec_str = f"OK. Result rows (preview): {rows_str[:300]}"
+                        rows = str(er[0])[:280]
+                        exec_str = f"OK. Rows preview: {rows}" if rows.strip() and rows.strip() != "[]" else "OK. (no rows returned)"
+                else:
+                    exec_str = f"{str(er[0])[:200]}" if er[1] else "OK"
                 return build_prompt_chat(sample, t_, exec_result_str=exec_str)
             futs = [exe.submit(score_via_vllm, args.selector_host, _make_prompt(i, t)) for i, t in enumerate(traj)]
             for f in futs: