Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 0.0000 ± 0.0000
total_entries: 6.0000 ± 0.0000
correct_matches: 0.0000 ± 0.0000
average_execution_time: 1.0523 ± 0.0157
metrics_by_complexity:
  complex:
    count: 1.8000 ± 0.4472
    correct_matches: 0.0000 ± 0.0000
    pv_matched: 1.0000 ± 0.0000
    pv_mismatched: 5.0000 ± 0.0000
    pv_match_rate: 16.6667 ± 0.0000
    timing_matched: 0.0000 ± 0.0000
    timing_mismatched: 6.0000 ± 0.0000
    timing_match_rate: 0.0000 ± 0.0000
    temp_matched: 1.0000 ± 0.0000
    temp_mismatched: 0.8000 ± 0.4472
    temp_match_rate: 60.0000 ± 22.3607
    full_matched: 0.0000 ± 0.0000
    full_mismatched: 6.0000 ± 0.0000
    full_match_rate: 0.0000 ± 0.0000
    accuracy: 0.0000 ± 0.0000
    average_best_codebleu:
      codebleu: 0.3466 ± 0.0107
      ngram_match_score: 0.0602 ± 0.0128
      weighted_ngram_match_score: 0.0680 ± 0.0087
      syntax_match_score: 0.7009 ± 0.0324
      dataflow_match_score: 0.5571 ± 0.0319
    average_best_levenshtein: 140.0000 ± 4.4721
    average_best_normalized_levenshtein: 0.4770 ± 0.0150
    average_execution_time: 1.1907 ± 0.0604
==================================================
