Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 85.4497 ± 0.4582
total_entries: 126.0000 ± 0.0000
correct_matches: 107.6667 ± 0.5774
average_execution_time: 1.1828 ± 0.1194
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 105.0000 ± 0.0000
    exact_code_matches: 100.0000 ± 0.0000
    exact_code_match_rate: 83.3333 ± 0.0000
    pv_exact_matched: 105.0000 ± 0.0000
    pv_exact_mismatched: 15.0000 ± 0.0000
    pv_exact_match_rate: 87.5000 ± 0.0000
    average_pv_match_rate: 93.7500 ± 0.0000
    average_pv_mismatch_rate: 6.2500 ± 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: 105.0000 ± 0.0000
    full_mismatched: 15.0000 ± 0.0000
    full_match_rate: 87.5000 ± 0.0000
    accuracy: 87.5000 ± 0.0000
    average_timing_score: 1.0000 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9500 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.6017 ± 0.0017
      comb_7_CodeBLEU: 0.8336 ± 0.0007
      dataflow_match_score: 0.1500 ± 0.0000
      ngram_match_score: 0.2605 ± 0.0033
      syntax_match_score: 0.9764 ± 0.0000
      weighted_ngram_match_score: 0.1699 ± 0.0036
    average_best_levenshtein: 2.8167 ± 0.0000
    average_best_normalized_levenshtein: 0.0555 ± 0.0000
    average_inference_time: 1.1548 ± 0.1239
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 2.6667 ± 0.5774
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 16.6667 ± 0.0000
    pv_exact_matched: 2.6667 ± 0.5774
    pv_exact_mismatched: 3.3333 ± 0.5774
    pv_exact_match_rate: 44.4444 ± 9.6225
    average_pv_match_rate: 85.8663 ± 1.4804
    average_pv_mismatch_rate: 14.1337 ± 1.4804
    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: 2.6667 ± 0.5774
    full_mismatched: 3.3333 ± 0.5774
    full_match_rate: 44.4444 ± 9.6225
    accuracy: 44.4444 ± 9.6225
    average_timing_score: 0.9199 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.8709 ± 0.0119
    average_best_codebleu:
      average_CodeBLEU: 0.6847 ± 0.0057
      comb_7_CodeBLEU: 0.7560 ± 0.0032
      dataflow_match_score: 0.7655 ± 0.0000
      ngram_match_score: 0.5395 ± 0.0127
      syntax_match_score: 0.8415 ± 0.0030
      weighted_ngram_match_score: 0.5922 ± 0.0071
    average_best_levenshtein: 31.0556 ± 5.0037
    average_best_normalized_levenshtein: 0.1735 ± 0.0188
    average_inference_time: 1.7430 ± 0.0858
==================================================
