Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 81.7460 ± 0.0000
total_entries: 126.0000 ± 0.0000
correct_matches: 103.0000 ± 0.0000
average_execution_time: 1.0300 ± 0.0854
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 101.0000 ± 0.0000
    exact_code_matches: 90.0000 ± 0.0000
    exact_code_match_rate: 75.0000 ± 0.0000
    pv_exact_matched: 101.0000 ± 0.0000
    pv_exact_mismatched: 19.0000 ± 0.0000
    pv_exact_match_rate: 84.1667 ± 0.0000
    average_pv_match_rate: 90.9722 ± 0.0000
    average_pv_mismatch_rate: 9.0278 ± 0.0000
    timing_matched: 119.0000 ± 0.0000
    timing_mismatched: 1.0000 ± 0.0000
    timing_match_rate: 99.1667 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 101.0000 ± 0.0000
    full_mismatched: 19.0000 ± 0.0000
    full_match_rate: 84.1667 ± 0.0000
    accuracy: 84.1667 ± 0.0000
    average_timing_score: 0.9916 ± 0.0001
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9261 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5671 ± 0.0000
      comb_7_CodeBLEU: 0.8172 ± 0.0000
      dataflow_match_score: 0.1000 ± 0.0000
      ngram_match_score: 0.1941 ± 0.0000
      syntax_match_score: 0.9678 ± 0.0000
      weighted_ngram_match_score: 0.1067 ± 0.0000
    average_best_levenshtein: 3.7583 ± 0.0000
    average_best_normalized_levenshtein: 0.0917 ± 0.0000
    average_inference_time: 1.0390 ± 0.0747
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 2.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 16.6667 ± 0.0000
    pv_exact_matched: 2.0000 ± 0.0000
    pv_exact_mismatched: 4.0000 ± 0.0000
    pv_exact_match_rate: 33.3333 ± 0.0000
    average_pv_match_rate: 79.3349 ± 1.9245
    average_pv_mismatch_rate: 20.6651 ± 1.9245
    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: 2.0000 ± 0.0000
    full_mismatched: 4.0000 ± 0.0000
    full_match_rate: 33.3333 ± 0.0000
    accuracy: 33.3333 ± 0.0000
    average_timing_score: 0.9100 ± 0.0017
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.8167 ± 0.0150
    average_best_codebleu:
      average_CodeBLEU: 0.5046 ± 0.0599
      comb_7_CodeBLEU: 0.6314 ± 0.0560
      dataflow_match_score: 0.5155 ± 0.0615
      ngram_match_score: 0.2860 ± 0.0673
      syntax_match_score: 0.7497 ± 0.0451
      weighted_ngram_match_score: 0.3007 ± 0.0658
    average_best_levenshtein: 58.3889 ± 12.6055
    average_best_normalized_levenshtein: 0.3982 ± 0.0521
    average_inference_time: 0.8491 ± 0.3308
==================================================
