Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 73.9130 ± 0.0000
total_entries: 138.0000 ± 0.0000
correct_matches: 102.0000 ± 0.0000
average_execution_time: 1.4591 ± 0.0232
metrics_by_complexity:
  simple:
    count: 118.0000 ± 0.0000
    full_matches: 96.0000 ± 0.0000
    exact_code_matches: 92.0000 ± 0.0000
    exact_code_match_rate: 77.9661 ± 0.0000
    pv_exact_matched: 97.0000 ± 0.0000
    pv_exact_mismatched: 21.0000 ± 0.0000
    pv_exact_match_rate: 82.2034 ± 0.0000
    average_pv_match_rate: 84.3220 ± 0.0000
    average_pv_mismatch_rate: 15.6780 ± 0.0000
    timing_matched: 101.0000 ± 0.0000
    timing_mismatched: 17.0000 ± 0.0000
    timing_match_rate: 85.5932 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 96.0000 ± 0.0000
    full_mismatched: 22.0000 ± 0.0000
    full_match_rate: 81.3559 ± 0.0000
    accuracy: 81.3559 ± 0.0000
    average_timing_score: 0.8599 ± 0.0002
    average_temp_score: 0.9492 ± 0.0000
    average_full_score: 0.8465 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5328 ± 0.0000
      comb_7_CodeBLEU: 0.7787 ± 0.0000
      dataflow_match_score: 0.0932 ± 0.0000
      ngram_match_score: 0.1712 ± 0.0000
      syntax_match_score: 0.8852 ± 0.0000
      weighted_ngram_match_score: 0.0749 ± 0.0000
    average_best_levenshtein: 3.1780 ± 0.0000
    average_best_normalized_levenshtein: 0.0941 ± 0.0000
    average_inference_time: 0.7525 ± 0.0265
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 6.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 5.0000 ± 0.0000
    pv_exact_matched: 7.0000 ± 0.0000
    pv_exact_mismatched: 13.0000 ± 0.0000
    pv_exact_match_rate: 35.0000 ± 0.0000
    average_pv_match_rate: 54.2444 ± 0.2887
    average_pv_mismatch_rate: 45.7556 ± 0.2887
    timing_matched: 12.3333 ± 0.5774
    timing_mismatched: 7.6667 ± 0.5774
    timing_match_rate: 61.6667 ± 2.8868
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 6.0000 ± 0.0000
    full_mismatched: 14.0000 ± 0.0000
    full_match_rate: 30.0000 ± 0.0000
    accuracy: 30.0000 ± 0.0000
    average_timing_score: 0.8225 ± 0.0071
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.5847 ± 0.0009
    average_best_codebleu:
      average_CodeBLEU: 0.3932 ± 0.0000
      comb_7_CodeBLEU: 0.5268 ± 0.0000
      dataflow_match_score: 0.6174 ± 0.0000
      ngram_match_score: 0.1603 ± 0.0000
      syntax_match_score: 0.6142 ± 0.0000
      weighted_ngram_match_score: 0.1810 ± 0.0000
    average_best_levenshtein: 143.2000 ± 0.0000
    average_best_normalized_levenshtein: 0.4422 ± 0.0000
    average_inference_time: 5.6279 ± 0.0136
==================================================
