Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 66.9118 ± 0.0000
total_entries: 136.0000 ± 0.0000
correct_matches: 91.0000 ± 0.0000
average_execution_time: 2.2215 ± 0.1972
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 86.0000 ± 0.0000
    exact_code_matches: 80.0000 ± 0.0000
    exact_code_match_rate: 68.9655 ± 0.0000
    pv_exact_matched: 86.0000 ± 0.0000
    pv_exact_mismatched: 30.0000 ± 0.0000
    pv_exact_match_rate: 74.1379 ± 0.0000
    average_pv_match_rate: 75.8621 ± 0.0000
    average_pv_mismatch_rate: 24.1379 ± 0.0000
    timing_matched: 90.0000 ± 0.0000
    timing_mismatched: 26.0000 ± 0.0000
    timing_match_rate: 77.5862 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 86.0000 ± 0.0000
    full_mismatched: 30.0000 ± 0.0000
    full_match_rate: 74.1379 ± 0.0000
    accuracy: 74.1379 ± 0.0000
    average_timing_score: 0.7758 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.7621 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5415 ± 0.0000
      comb_7_CodeBLEU: 0.7987 ± 0.0000
      dataflow_match_score: 0.1207 ± 0.0000
      ngram_match_score: 0.1550 ± 0.0000
      syntax_match_score: 0.9402 ± 0.0000
      weighted_ngram_match_score: 0.0709 ± 0.0000
    average_best_levenshtein: 2.6379 ± 0.0000
    average_best_normalized_levenshtein: 0.1017 ± 0.0000
    average_inference_time: 1.3535 ± 0.0333
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 5.0000 ± 0.0000
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 5.0000 ± 0.0000
    pv_exact_mismatched: 15.0000 ± 0.0000
    pv_exact_match_rate: 25.0000 ± 0.0000
    average_pv_match_rate: 43.9651 ± 1.5940
    average_pv_mismatch_rate: 56.0349 ± 1.5940
    timing_matched: 12.6667 ± 0.5774
    timing_mismatched: 7.3333 ± 0.5774
    timing_match_rate: 63.3333 ± 2.8868
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 5.0000 ± 0.0000
    full_mismatched: 15.0000 ± 0.0000
    full_match_rate: 25.0000 ± 0.0000
    accuracy: 25.0000 ± 0.0000
    average_timing_score: 0.6945 ± 0.0267
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.4860 ± 0.0181
    average_best_codebleu:
      average_CodeBLEU: 0.4216 ± 0.0020
      comb_7_CodeBLEU: 0.5502 ± 0.0033
      dataflow_match_score: 0.6115 ± 0.0002
      ngram_match_score: 0.1950 ± 0.0000
      syntax_match_score: 0.6604 ± 0.0083
      weighted_ngram_match_score: 0.2195 ± 0.0001
    average_best_levenshtein: 187.6500 ± 1.0112
    average_best_normalized_levenshtein: 0.4506 ± 0.0009
    average_inference_time: 7.2559 ± 1.1485
==================================================
