Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 85.7843 ± 0.4245
total_entries: 136.0000 ± 0.0000
correct_matches: 116.6667 ± 0.5774
average_execution_time: 0.7899 ± 0.1484
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 113.0000 ± 0.0000
    exact_code_matches: 108.0000 ± 0.0000
    exact_code_match_rate: 93.1034 ± 0.0000
    pv_exact_matched: 114.0000 ± 0.0000
    pv_exact_mismatched: 2.0000 ± 0.0000
    pv_exact_match_rate: 98.2759 ± 0.0000
    average_pv_match_rate: 98.7069 ± 0.0000
    average_pv_mismatch_rate: 1.2931 ± 0.0000
    timing_matched: 114.0000 ± 0.0000
    timing_mismatched: 2.0000 ± 0.0000
    timing_match_rate: 98.2759 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 113.0000 ± 0.0000
    full_mismatched: 3.0000 ± 0.0000
    full_match_rate: 97.4138 ± 0.0000
    accuracy: 97.4138 ± 0.0000
    average_timing_score: 0.9871 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9871 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5951 ± 0.0037
      comb_7_CodeBLEU: 0.8378 ± 0.0015
      dataflow_match_score: 0.2586 ± 0.0086
      ngram_match_score: 0.2395 ± 0.0071
      syntax_match_score: 0.9990 ± 0.0000
      weighted_ngram_match_score: 0.1420 ± 0.0078
    average_best_levenshtein: 0.4741 ± 0.0000
    average_best_normalized_levenshtein: 0.0150 ± 0.0000
    average_inference_time: 0.4144 ± 0.0602
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 3.6667 ± 0.5774
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 4.0000 ± 1.0000
    pv_exact_mismatched: 16.0000 ± 1.0000
    pv_exact_match_rate: 20.0000 ± 5.0000
    average_pv_match_rate: 51.0882 ± 3.2454
    average_pv_mismatch_rate: 48.9118 ± 3.2454
    timing_matched: 13.3333 ± 1.5275
    timing_mismatched: 6.6667 ± 1.5275
    timing_match_rate: 66.6667 ± 7.6376
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 3.6667 ± 0.5774
    full_mismatched: 16.3333 ± 0.5774
    full_match_rate: 18.3333 ± 2.8868
    accuracy: 18.3333 ± 2.8868
    average_timing_score: 0.7941 ± 0.0384
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.5464 ± 0.0280
    average_best_codebleu:
      average_CodeBLEU: 0.3920 ± 0.0102
      comb_7_CodeBLEU: 0.5104 ± 0.0119
      dataflow_match_score: 0.4806 ± 0.0077
      ngram_match_score: 0.1837 ± 0.0077
      syntax_match_score: 0.6480 ± 0.0184
      weighted_ngram_match_score: 0.2057 ± 0.0075
    average_best_levenshtein: 141.2667 ± 1.6065
    average_best_normalized_levenshtein: 0.4245 ± 0.0041
    average_inference_time: 2.9677 ± 0.6641
==================================================
