Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 75.0000 ± 0.0000
total_entries: 136.0000 ± 0.0000
correct_matches: 102.0000 ± 0.0000
average_execution_time: 1.3817 ± 0.4918
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 96.0000 ± 0.0000
    exact_code_matches: 92.0000 ± 0.0000
    exact_code_match_rate: 79.3103 ± 0.0000
    pv_exact_matched: 97.0000 ± 0.0000
    pv_exact_mismatched: 19.0000 ± 0.0000
    pv_exact_match_rate: 83.6207 ± 0.0000
    average_pv_match_rate: 85.3448 ± 0.0000
    average_pv_mismatch_rate: 14.6552 ± 0.0000
    timing_matched: 100.0000 ± 0.0000
    timing_mismatched: 16.0000 ± 0.0000
    timing_match_rate: 86.2069 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 96.0000 ± 0.0000
    full_mismatched: 20.0000 ± 0.0000
    full_match_rate: 82.7586 ± 0.0000
    accuracy: 82.7586 ± 0.0000
    average_timing_score: 0.8659 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.8559 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5369 ± 0.0000
      comb_7_CodeBLEU: 0.7847 ± 0.0000
      dataflow_match_score: 0.1207 ± 0.0000
      ngram_match_score: 0.1728 ± 0.0000
      syntax_match_score: 0.8996 ± 0.0000
      weighted_ngram_match_score: 0.0751 ± 0.0000
    average_best_levenshtein: 2.6121 ± 0.0000
    average_best_normalized_levenshtein: 0.0804 ± 0.0000
    average_inference_time: 0.9570 ± 0.3842
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 6.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 5.0000 ± 0.0000
    pv_exact_matched: 7.0000 ± 0.0000
    pv_exact_mismatched: 13.0000 ± 0.0000
    pv_exact_match_rate: 35.0000 ± 0.0000
    average_pv_match_rate: 53.3267 ± 0.2507
    average_pv_mismatch_rate: 46.6733 ± 0.2507
    timing_matched: 11.0000 ± 1.0000
    timing_mismatched: 9.0000 ± 1.0000
    timing_match_rate: 55.0000 ± 5.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 6.0000 ± 0.0000
    full_mismatched: 14.0000 ± 0.0000
    full_match_rate: 30.0000 ± 0.0000
    accuracy: 30.0000 ± 0.0000
    average_timing_score: 0.7366 ± 0.0311
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.5620 ± 0.0059
    average_best_codebleu:
      average_CodeBLEU: 0.4015 ± 0.0010
      comb_7_CodeBLEU: 0.5375 ± 0.0015
      dataflow_match_score: 0.6214 ± 0.0000
      ngram_match_score: 0.1650 ± 0.0003
      syntax_match_score: 0.6347 ± 0.0037
      weighted_ngram_match_score: 0.1850 ± 0.0002
    average_best_levenshtein: 139.6333 ± 0.1443
    average_best_normalized_levenshtein: 0.4298 ± 0.0002
    average_inference_time: 3.8448 ± 1.1186
==================================================
