Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 80.1587 ± 0.0000
total_entries: 126.0000 ± 0.0000
correct_matches: 101.0000 ± 0.0000
average_execution_time: 1.1329 ± 0.0857
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 97.0000 ± 0.0000
    exact_code_matches: 78.0000 ± 0.0000
    exact_code_match_rate: 65.0000 ± 0.0000
    pv_exact_matched: 100.0000 ± 0.0000
    pv_exact_mismatched: 20.0000 ± 0.0000
    pv_exact_match_rate: 83.3333 ± 0.0000
    average_pv_match_rate: 90.9722 ± 0.0000
    average_pv_mismatch_rate: 9.0278 ± 0.0000
    timing_matched: 117.0000 ± 0.0000
    timing_mismatched: 3.0000 ± 0.0000
    timing_match_rate: 97.5000 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 97.0000 ± 0.0000
    full_mismatched: 23.0000 ± 0.0000
    full_match_rate: 80.8333 ± 0.0000
    accuracy: 80.8333 ± 0.0000
    average_timing_score: 0.9921 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9262 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5444 ± 0.0000
      comb_7_CodeBLEU: 0.7873 ± 0.0000
      dataflow_match_score: 0.1000 ± 0.0000
      ngram_match_score: 0.1778 ± 0.0000
      syntax_match_score: 0.8985 ± 0.0000
      weighted_ngram_match_score: 0.1011 ± 0.0000
    average_best_levenshtein: 4.0417 ± 0.0000
    average_best_normalized_levenshtein: 0.1057 ± 0.0000
    average_inference_time: 1.1485 ± 0.0765
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 4.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 16.6667 ± 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: 92.2547 ± 0.0000
    average_pv_mismatch_rate: 7.7453 ± 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: 0.8983 ± 0.0006
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9177 ± 0.0001
    average_best_codebleu:
      average_CodeBLEU: 0.7115 ± 0.0003
      comb_7_CodeBLEU: 0.8097 ± 0.0001
      dataflow_match_score: 0.8980 ± 0.0000
      ngram_match_score: 0.5024 ± 0.0006
      syntax_match_score: 0.8523 ± 0.0000
      weighted_ngram_match_score: 0.5933 ± 0.0005
    average_best_levenshtein: 22.5000 ± 0.2887
    average_best_normalized_levenshtein: 0.1482 ± 0.0014
    average_inference_time: 0.8216 ± 0.3645
==================================================
