Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 77.7778 ± 0.0000
total_entries: 126.0000 ± 0.0000
correct_matches: 98.0000 ± 0.0000
average_execution_time: 1.0021 ± 0.0332
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 96.0000 ± 0.0000
    exact_code_matches: 79.0000 ± 0.0000
    exact_code_match_rate: 65.8333 ± 0.0000
    pv_exact_matched: 99.0000 ± 0.0000
    pv_exact_mismatched: 21.0000 ± 0.0000
    pv_exact_match_rate: 82.5000 ± 0.0000
    average_pv_match_rate: 90.1389 ± 0.0000
    average_pv_mismatch_rate: 9.8611 ± 0.0000
    timing_matched: 116.0000 ± 0.0000
    timing_mismatched: 4.0000 ± 0.0000
    timing_match_rate: 96.6667 ± 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: 24.0000 ± 0.0000
    full_match_rate: 80.0000 ± 0.0000
    accuracy: 80.0000 ± 0.0000
    average_timing_score: 0.9838 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9179 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5431 ± 0.0000
      comb_7_CodeBLEU: 0.7848 ± 0.0000
      dataflow_match_score: 0.1000 ± 0.0000
      ngram_match_score: 0.1793 ± 0.0000
      syntax_match_score: 0.8919 ± 0.0000
      weighted_ngram_match_score: 0.1014 ± 0.0000
    average_best_levenshtein: 4.0083 ± 0.0000
    average_best_normalized_levenshtein: 0.1034 ± 0.0000
    average_inference_time: 1.0222 ± 0.0347
  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: 33.3333 ± 0.0000
    average_pv_mismatch_rate: 66.6667 ± 0.0000
    timing_matched: 2.0000 ± 0.0000
    timing_mismatched: 4.0000 ± 0.0000
    timing_match_rate: 33.3333 ± 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: 0.2971 ± 0.0000
    average_temp_score: 0.3333 ± 0.0000
    average_full_score: 0.3261 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.7246 ± 0.0000
      comb_7_CodeBLEU: 0.8308 ± 0.0000
      dataflow_match_score: 0.9510 ± 0.0000
      ngram_match_score: 0.5021 ± 0.0000
      syntax_match_score: 0.8523 ± 0.0000
      weighted_ngram_match_score: 0.5930 ± 0.0000
    average_best_levenshtein: 22.6667 ± 0.0000
    average_best_normalized_levenshtein: 0.1490 ± 0.0000
    average_inference_time: 0.6004 ± 0.0033
==================================================
