Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 72.2222 ± 0.0000
total_entries: 126.0000 ± 0.0000
correct_matches: 91.0000 ± 0.0000
average_execution_time: 1.9323 ± 0.8051
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 87.0000 ± 0.0000
    exact_code_matches: 80.0000 ± 0.0000
    exact_code_match_rate: 66.6667 ± 0.0000
    pv_exact_matched: 87.0000 ± 0.0000
    pv_exact_mismatched: 33.0000 ± 0.0000
    pv_exact_match_rate: 72.5000 ± 0.0000
    average_pv_match_rate: 75.0000 ± 0.0000
    average_pv_mismatch_rate: 25.0000 ± 0.0000
    timing_matched: 93.0000 ± 0.0000
    timing_mismatched: 27.0000 ± 0.0000
    timing_match_rate: 77.5000 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 87.0000 ± 0.0000
    full_mismatched: 33.0000 ± 0.0000
    full_match_rate: 72.5000 ± 0.0000
    accuracy: 72.5000 ± 0.0000
    average_timing_score: 0.7746 ± 0.0000
    average_temp_score: 0.9083 ± 0.0000
    average_full_score: 0.7549 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5364 ± 0.0000
      comb_7_CodeBLEU: 0.7906 ± 0.0000
      dataflow_match_score: 0.1000 ± 0.0000
      ngram_match_score: 0.1529 ± 0.0000
      syntax_match_score: 0.9200 ± 0.0000
      weighted_ngram_match_score: 0.0726 ± 0.0000
    average_best_levenshtein: 3.4917 ± 0.0000
    average_best_normalized_levenshtein: 0.1208 ± 0.0000
    average_inference_time: 1.7430 ± 0.7109
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 4.0000 ± 0.0000
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 4.0000 ± 0.0000
    pv_exact_mismatched: 2.0000 ± 0.0000
    pv_exact_match_rate: 66.6667 ± 0.0000
    average_pv_match_rate: 84.9462 ± 0.0000
    average_pv_mismatch_rate: 15.0538 ± 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: 4.0000 ± 0.0000
    full_mismatched: 2.0000 ± 0.0000
    full_match_rate: 66.6667 ± 0.0000
    accuracy: 66.6667 ± 0.0000
    average_timing_score: 0.9685 ± 0.0004
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.8733 ± 0.0001
    average_best_codebleu:
      average_CodeBLEU: 0.6714 ± 0.0085
      comb_7_CodeBLEU: 0.7762 ± 0.0049
      dataflow_match_score: 0.8273 ± 0.0025
      ngram_match_score: 0.4544 ± 0.0127
      syntax_match_score: 0.8649 ± 0.0024
      weighted_ngram_match_score: 0.5389 ± 0.0164
    average_best_levenshtein: 31.7778 ± 0.1925
    average_best_normalized_levenshtein: 0.1675 ± 0.0018
    average_inference_time: 5.7165 ± 2.7255
==================================================
