Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 81.6176 ± 0.0000
total_entries: 136.0000 ± 0.0000
correct_matches: 111.0000 ± 0.0000
average_execution_time: 1.1725 ± 0.1256
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 109.0000 ± 0.0000
    exact_code_matches: 99.6667 ± 0.5774
    exact_code_match_rate: 85.9195 ± 0.4977
    pv_exact_matched: 109.0000 ± 0.0000
    pv_exact_mismatched: 7.0000 ± 0.0000
    pv_exact_match_rate: 93.9655 ± 0.0000
    average_pv_match_rate: 95.4023 ± 0.0000
    average_pv_mismatch_rate: 4.5977 ± 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: 109.0000 ± 0.0000
    full_mismatched: 7.0000 ± 0.0000
    full_match_rate: 93.9655 ± 0.0000
    accuracy: 93.9655 ± 0.0000
    average_timing_score: 0.9827 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9598 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5950 ± 0.0013
      comb_7_CodeBLEU: 0.8347 ± 0.0017
      dataflow_match_score: 0.2155 ± 0.0000
      ngram_match_score: 0.2417 ± 0.0009
      syntax_match_score: 0.9891 ± 0.0040
      weighted_ngram_match_score: 0.1490 ± 0.0002
    average_best_levenshtein: 1.4023 ± 0.0100
    average_best_normalized_levenshtein: 0.0515 ± 0.0009
    average_inference_time: 1.0856 ± 0.0718
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 2.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 5.0000 ± 0.0000
    pv_exact_matched: 2.0000 ± 0.0000
    pv_exact_mismatched: 18.0000 ± 0.0000
    pv_exact_match_rate: 10.0000 ± 0.0000
    average_pv_match_rate: 44.9161 ± 0.6900
    average_pv_mismatch_rate: 55.0839 ± 0.6900
    timing_matched: 11.6667 ± 0.5774
    timing_mismatched: 8.3333 ± 0.5774
    timing_match_rate: 58.3333 ± 2.8868
    temp_matched: 4.3333 ± 0.5774
    temp_mismatched: 4.6667 ± 0.5774
    temp_match_rate: 48.1481 ± 6.4150
    full_matched: 2.0000 ± 0.0000
    full_mismatched: 18.0000 ± 0.0000
    full_match_rate: 10.0000 ± 0.0000
    accuracy: 10.0000 ± 0.0000
    average_timing_score: 0.7595 ± 0.0122
    average_temp_score: 0.7737 ± 0.0286
    average_full_score: 0.5211 ± 0.0097
    average_best_codebleu:
      average_CodeBLEU: 0.3631 ± 0.0016
      comb_7_CodeBLEU: 0.4822 ± 0.0027
      dataflow_match_score: 0.4110 ± 0.0054
      ngram_match_score: 0.1603 ± 0.0026
      syntax_match_score: 0.5623 ± 0.0022
      weighted_ngram_match_score: 0.1688 ± 0.0034
    average_best_levenshtein: 158.7500 ± 6.5962
    average_best_normalized_levenshtein: 0.5061 ± 0.0043
    average_inference_time: 1.6764 ± 0.4609
==================================================
