Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 87.9902 ± 0.4245
total_entries: 136.0000 ± 0.0000
correct_matches: 119.6667 ± 0.5774
average_execution_time: 26.2296 ± 0.2915
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 113.0000 ± 0.0000
    exact_code_matches: 109.0000 ± 0.0000
    exact_code_match_rate: 93.9655 ± 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: 99.2816 ± 0.0000
    average_pv_mismatch_rate: 0.7184 ± 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: 0.9900 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9922 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.6204 ± 0.0000
      comb_7_CodeBLEU: 0.8452 ± 0.0000
      dataflow_match_score: 0.2845 ± 0.0000
      ngram_match_score: 0.2935 ± 0.0000
      syntax_match_score: 0.9903 ± 0.0000
      weighted_ngram_match_score: 0.1976 ± 0.0000
    average_best_levenshtein: 0.7931 ± 0.0000
    average_best_normalized_levenshtein: 0.0158 ± 0.0000
    average_inference_time: 28.8275 ± 0.0292
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 6.6667 ± 0.5774
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 10.0000 ± 0.0000
    pv_exact_mismatched: 10.0000 ± 0.0000
    pv_exact_match_rate: 50.0000 ± 0.0000
    average_pv_match_rate: 81.4779 ± 0.1443
    average_pv_mismatch_rate: 18.5221 ± 0.1443
    timing_matched: 12.6667 ± 0.5774
    timing_mismatched: 7.3333 ± 0.5774
    timing_match_rate: 63.3333 ± 2.8868
    temp_matched: 8.0000 ± 0.0000
    temp_mismatched: 1.0000 ± 0.0000
    temp_match_rate: 88.8889 ± 0.0000
    full_matched: 6.6667 ± 0.5774
    full_mismatched: 13.3333 ± 0.5774
    full_match_rate: 33.3333 ± 2.8868
    accuracy: 33.3333 ± 2.8868
    average_timing_score: 0.7360 ± 0.0056
    average_temp_score: 0.9802 ± 0.0015
    average_full_score: 0.8004 ± 0.0018
    average_best_codebleu:
      average_CodeBLEU: 0.4587 ± 0.0040
      comb_7_CodeBLEU: 0.5770 ± 0.0024
      dataflow_match_score: 0.5536 ± 0.0036
      ngram_match_score: 0.2392 ± 0.0087
      syntax_match_score: 0.7082 ± 0.0005
      weighted_ngram_match_score: 0.2837 ± 0.0056
    average_best_levenshtein: 156.3167 ± 3.6250
    average_best_normalized_levenshtein: 0.4189 ± 0.0055
    average_inference_time: 11.1618 ± 2.1209
==================================================
