Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 87.0098 ± 0.4245
total_entries: 136.0000 ± 0.0000
correct_matches: 118.3333 ± 0.5774
average_execution_time: 1.8016 ± 0.2937
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 113.3333 ± 0.5774
    exact_code_matches: 102.6667 ± 0.5774
    exact_code_match_rate: 88.5057 ± 0.4977
    pv_exact_matched: 115.3333 ± 0.5774
    pv_exact_mismatched: 0.6667 ± 0.5774
    pv_exact_match_rate: 99.4253 ± 0.4977
    average_pv_match_rate: 99.7126 ± 0.2489
    average_pv_mismatch_rate: 0.2874 ± 0.2489
    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: 113.3333 ± 0.5774
    full_mismatched: 2.6667 ± 0.5774
    full_match_rate: 97.7011 ± 0.4977
    accuracy: 97.7011 ± 0.4977
    average_timing_score: 0.9908 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9959 ± 0.0020
    average_best_codebleu:
      average_CodeBLEU: 0.6020 ± 0.0040
      comb_7_CodeBLEU: 0.8385 ± 0.0017
      dataflow_match_score: 0.2874 ± 0.0050
      ngram_match_score: 0.2500 ± 0.0077
      syntax_match_score: 0.9923 ± 0.0005
      weighted_ngram_match_score: 0.1659 ± 0.0083
    average_best_levenshtein: 2.1810 ± 0.2160
    average_best_normalized_levenshtein: 0.0293 ± 0.0018
    average_inference_time: 0.8065 ± 0.1380
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 5.0000 ± 0.0000
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 5.0000 ± 0.0000
    pv_exact_mismatched: 15.0000 ± 0.0000
    pv_exact_match_rate: 25.0000 ± 0.0000
    average_pv_match_rate: 45.0071 ± 2.5615
    average_pv_mismatch_rate: 54.9929 ± 2.5615
    timing_matched: 14.3333 ± 0.5774
    timing_mismatched: 5.6667 ± 0.5774
    timing_match_rate: 71.6667 ± 2.8868
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 5.0000 ± 0.0000
    full_mismatched: 15.0000 ± 0.0000
    full_match_rate: 25.0000 ± 0.0000
    accuracy: 25.0000 ± 0.0000
    average_timing_score: 0.7730 ± 0.0067
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.4974 ± 0.0147
    average_best_codebleu:
      average_CodeBLEU: 0.4107 ± 0.0039
      comb_7_CodeBLEU: 0.5423 ± 0.0026
      dataflow_match_score: 0.5987 ± 0.0013
      ngram_match_score: 0.1801 ± 0.0061
      syntax_match_score: 0.6616 ± 0.0022
      weighted_ngram_match_score: 0.2023 ± 0.0061
    average_best_levenshtein: 191.6167 ± 7.3997
    average_best_normalized_levenshtein: 0.4559 ± 0.0016
    average_inference_time: 7.5733 ± 1.3675
==================================================
