Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 89.7059 ± 1.9454
total_entries: 136.0000 ± 0.0000
correct_matches: 122.0000 ± 2.6458
average_execution_time: 6.6888 ± 0.2028
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 110.6667 ± 1.1547
    exact_code_matches: 102.3333 ± 1.5275
    exact_code_match_rate: 88.2184 ± 1.3168
    pv_exact_matched: 113.6667 ± 1.1547
    pv_exact_mismatched: 2.3333 ± 1.1547
    pv_exact_match_rate: 97.9885 ± 0.9954
    average_pv_match_rate: 98.5722 ± 0.9877
    average_pv_mismatch_rate: 1.4278 ± 0.9877
    timing_matched: 111.0000 ± 1.0000
    timing_mismatched: 5.0000 ± 1.0000
    timing_match_rate: 95.6897 ± 0.8621
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 110.6667 ± 1.1547
    full_mismatched: 5.3333 ± 1.1547
    full_match_rate: 95.4023 ± 0.9954
    accuracy: 95.4023 ± 0.9954
    average_timing_score: 0.9744 ± 0.0093
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9835 ± 0.0096
    average_best_codebleu:
      average_CodeBLEU: 0.6326 ± 0.0055
      comb_7_CodeBLEU: 0.8499 ± 0.0036
      dataflow_match_score: 0.2960 ± 0.0100
      ngram_match_score: 0.3108 ± 0.0092
      syntax_match_score: 0.9894 ± 0.0046
      weighted_ngram_match_score: 0.2302 ± 0.0085
    average_best_levenshtein: 1.7500 ± 0.3511
    average_best_normalized_levenshtein: 0.0259 ± 0.0046
    average_inference_time: 5.7371 ± 0.0645
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 11.3333 ± 1.5275
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 12.3333 ± 1.5275
    pv_exact_mismatched: 7.6667 ± 1.5275
    pv_exact_match_rate: 61.6667 ± 7.6376
    average_pv_match_rate: 90.0652 ± 2.7424
    average_pv_mismatch_rate: 9.9348 ± 2.7424
    timing_matched: 17.6667 ± 0.5774
    timing_mismatched: 2.3333 ± 0.5774
    timing_match_rate: 88.3333 ± 2.8868
    temp_matched: 7.3333 ± 0.5774
    temp_mismatched: 1.6667 ± 0.5774
    temp_match_rate: 81.4815 ± 6.4150
    full_matched: 11.3333 ± 1.5275
    full_mismatched: 8.6667 ± 1.5275
    full_match_rate: 56.6667 ± 7.6376
    accuracy: 56.6667 ± 7.6376
    average_timing_score: 0.8331 ± 0.0124
    average_temp_score: 0.9737 ± 0.0071
    average_full_score: 0.8866 ± 0.0238
    average_best_codebleu:
      average_CodeBLEU: 0.4537 ± 0.0059
      comb_7_CodeBLEU: 0.5776 ± 0.0104
      dataflow_match_score: 0.5077 ± 0.0188
      ngram_match_score: 0.2326 ± 0.0045
      syntax_match_score: 0.7125 ± 0.0153
      weighted_ngram_match_score: 0.2622 ± 0.0042
    average_best_levenshtein: 142.8167 ± 4.6047
    average_best_normalized_levenshtein: 0.4334 ± 0.0056
    average_inference_time: 12.2086 ± 1.1828
==================================================
