Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 92.8922 ± 0.4245
total_entries: 136.0000 ± 0.0000
correct_matches: 126.3333 ± 0.5774
average_execution_time: 1.1067 ± 0.0271
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 114.0000 ± 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: 99.5402 ± 0.0000
    average_pv_mismatch_rate: 0.4598 ± 0.0000
    timing_matched: 115.0000 ± 0.0000
    timing_mismatched: 1.0000 ± 0.0000
    timing_match_rate: 99.1379 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 114.0000 ± 0.0000
    full_mismatched: 2.0000 ± 0.0000
    full_match_rate: 98.2759 ± 0.0000
    accuracy: 98.2759 ± 0.0000
    average_timing_score: 0.9942 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9952 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.6321 ± 0.0000
      comb_7_CodeBLEU: 0.8523 ± 0.0000
      dataflow_match_score: 0.2845 ± 0.0000
      ngram_match_score: 0.3114 ± 0.0000
      syntax_match_score: 0.9982 ± 0.0000
      weighted_ngram_match_score: 0.2187 ± 0.0000
    average_best_levenshtein: 0.6379 ± 0.0000
    average_best_normalized_levenshtein: 0.0139 ± 0.0000
    average_inference_time: 0.9404 ± 0.0448
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 12.3333 ± 0.5774
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 5.0000 ± 0.0000
    pv_exact_matched: 13.0000 ± 0.0000
    pv_exact_mismatched: 7.0000 ± 0.0000
    pv_exact_match_rate: 65.0000 ± 0.0000
    average_pv_match_rate: 90.4644 ± 0.5753
    average_pv_mismatch_rate: 9.5356 ± 0.5753
    timing_matched: 15.3333 ± 0.5774
    timing_mismatched: 4.6667 ± 0.5774
    timing_match_rate: 76.6667 ± 2.8868
    temp_matched: 7.0000 ± 0.0000
    temp_mismatched: 1.0000 ± 0.0000
    temp_match_rate: 87.5000 ± 0.0000
    full_matched: 12.3333 ± 0.5774
    full_mismatched: 7.6667 ± 0.5774
    full_match_rate: 61.6667 ± 2.8868
    accuracy: 61.6667 ± 2.8868
    average_timing_score: 0.8449 ± 0.0006
    average_temp_score: 0.9831 ± 0.0033
    average_full_score: 0.8977 ± 0.0047
    average_best_codebleu:
      average_CodeBLEU: 0.4308 ± 0.0075
      comb_7_CodeBLEU: 0.5241 ± 0.0014
      dataflow_match_score: 0.4395 ± 0.0095
      ngram_match_score: 0.2672 ± 0.0186
      syntax_match_score: 0.6331 ± 0.0021
      weighted_ngram_match_score: 0.2833 ± 0.0197
    average_best_levenshtein: 126.4333 ± 0.9359
    average_best_normalized_levenshtein: 0.3937 ± 0.0023
    average_inference_time: 2.0710 ± 0.1081
==================================================
