Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 53.6765 ± 0.0000
total_entries: 136.0000 ± 0.0000
correct_matches: 73.0000 ± 0.0000
average_execution_time: 0.7698 ± 0.0907
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 71.0000 ± 0.0000
    exact_code_matches: 35.0000 ± 0.0000
    exact_code_match_rate: 30.1724 ± 0.0000
    pv_exact_matched: 71.0000 ± 0.0000
    pv_exact_mismatched: 45.0000 ± 0.0000
    pv_exact_match_rate: 61.2069 ± 0.0000
    average_pv_match_rate: 64.6552 ± 0.0000
    average_pv_mismatch_rate: 35.3448 ± 0.0000
    timing_matched: 79.0000 ± 0.0000
    timing_mismatched: 37.0000 ± 0.0000
    timing_match_rate: 68.1034 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 71.0000 ± 0.0000
    full_mismatched: 45.0000 ± 0.0000
    full_match_rate: 61.2069 ± 0.0000
    accuracy: 61.2069 ± 0.0000
    average_timing_score: 0.6788 ± 0.0000
    average_temp_score: 0.9052 ± 0.0000
    average_full_score: 0.6530 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.4697 ± 0.0000
      comb_7_CodeBLEU: 0.7030 ± 0.0000
      dataflow_match_score: 0.2241 ± 0.0000
      ngram_match_score: 0.1012 ± 0.0000
      syntax_match_score: 0.7172 ± 0.0000
      weighted_ngram_match_score: 0.0606 ± 0.0000
    average_best_levenshtein: 7.3793 ± 0.0000
    average_best_normalized_levenshtein: 0.2581 ± 0.0000
    average_inference_time: 0.6049 ± 0.0218
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 2.0000 ± 0.0000
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 3.3333 ± 0.5774
    pv_exact_mismatched: 16.6667 ± 0.5774
    pv_exact_match_rate: 16.6667 ± 2.8868
    average_pv_match_rate: 38.0585 ± 1.9339
    average_pv_mismatch_rate: 61.9415 ± 1.9339
    timing_matched: 8.6667 ± 0.5774
    timing_mismatched: 11.3333 ± 0.5774
    timing_match_rate: 43.3333 ± 2.8868
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 2.0000 ± 0.0000
    full_mismatched: 18.0000 ± 0.0000
    full_match_rate: 10.0000 ± 0.0000
    accuracy: 10.0000 ± 0.0000
    average_timing_score: 0.5248 ± 0.0091
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.3954 ± 0.0096
    average_best_codebleu:
      average_CodeBLEU: 0.3538 ± 0.0020
      comb_7_CodeBLEU: 0.4711 ± 0.0019
      dataflow_match_score: 0.5156 ± 0.0029
      ngram_match_score: 0.1470 ± 0.0016
      syntax_match_score: 0.5829 ± 0.0008
      weighted_ngram_match_score: 0.1696 ± 0.0027
    average_best_levenshtein: 146.3167 ± 0.4907
    average_best_normalized_levenshtein: 0.4734 ± 0.0004
    average_inference_time: 1.7262 ± 0.5709
==================================================
