Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 84.9206 ± 0.0000
total_entries: 126.0000 ± 0.0000
correct_matches: 107.0000 ± 0.0000
average_execution_time: 0.7194 ± 0.0578
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 104.0000 ± 0.0000
    exact_code_matches: 99.0000 ± 0.0000
    exact_code_match_rate: 82.5000 ± 0.0000
    pv_exact_matched: 104.0000 ± 0.0000
    pv_exact_mismatched: 16.0000 ± 0.0000
    pv_exact_match_rate: 86.6667 ± 0.0000
    average_pv_match_rate: 93.3333 ± 0.0000
    average_pv_mismatch_rate: 6.6667 ± 0.0000
    timing_matched: 120.0000 ± 0.0000
    timing_mismatched: 0.0000 ± 0.0000
    timing_match_rate: 100.0000 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 104.0000 ± 0.0000
    full_mismatched: 16.0000 ± 0.0000
    full_match_rate: 86.6667 ± 0.0000
    accuracy: 86.6667 ± 0.0000
    average_timing_score: 1.0000 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9467 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.6132 ± 0.0044
      comb_7_CodeBLEU: 0.8375 ± 0.0024
      dataflow_match_score: 0.1972 ± 0.0096
      ngram_match_score: 0.2813 ± 0.0076
      syntax_match_score: 0.9739 ± 0.0029
      weighted_ngram_match_score: 0.1977 ± 0.0082
    average_best_levenshtein: 3.0722 ± 0.0096
    average_best_normalized_levenshtein: 0.0587 ± 0.0003
    average_inference_time: 0.6931 ± 0.0630
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 3.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 16.6667 ± 0.0000
    pv_exact_matched: 3.0000 ± 0.0000
    pv_exact_mismatched: 3.0000 ± 0.0000
    pv_exact_match_rate: 50.0000 ± 0.0000
    average_pv_match_rate: 86.7210 ± 0.0000
    average_pv_mismatch_rate: 13.2790 ± 0.0000
    timing_matched: 6.0000 ± 0.0000
    timing_mismatched: 0.0000 ± 0.0000
    timing_match_rate: 100.0000 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 3.0000 ± 0.0000
    full_mismatched: 3.0000 ± 0.0000
    full_match_rate: 50.0000 ± 0.0000
    accuracy: 50.0000 ± 0.0000
    average_timing_score: 0.9195 ± 0.0005
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.8777 ± 0.0001
    average_best_codebleu:
      average_CodeBLEU: 0.6313 ± 0.0179
      comb_7_CodeBLEU: 0.7118 ± 0.0108
      dataflow_match_score: 0.7457 ± 0.0038
      ngram_match_score: 0.4857 ± 0.0259
      syntax_match_score: 0.7851 ± 0.0096
      weighted_ngram_match_score: 0.5088 ± 0.0353
    average_best_levenshtein: 37.7222 ± 2.1752
    average_best_normalized_levenshtein: 0.1887 ± 0.0125
    average_inference_time: 1.2446 ± 0.0466
==================================================
