Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 100.0000 ± 0.0000
total_entries: 2.0000 ± 0.0000
correct_matches: 2.0000 ± 0.0000
average_execution_time: 5.2532 ± 3.2826
metrics_by_complexity:
  simple:
    count: 1.0000 ± 0.0000
    full_matches: 1.0000 ± 0.0000
    pv_exact_matched: 1.0000 ± 0.0000
    pv_exact_mismatched: 0.0000 ± 0.0000
    pv_exact_match_rate: 100.0000 ± 0.0000
    average_pv_match_rate: 100.0000 ± 0.0000
    average_pv_mismatch_rate: 0.0000 ± 0.0000
    timing_matched: 1.0000 ± 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: 1.0000 ± 0.0000
    full_mismatched: 0.0000 ± 0.0000
    full_match_rate: 100.0000 ± 0.0000
    accuracy: 100.0000 ± 0.0000
    average_best_codebleu:
      codebleu: 0.6419 ± 0.0000
      ngram_match_score: 0.3162 ± 0.0000
      weighted_ngram_match_score: 0.2515 ± 0.0000
      syntax_match_score: 1.0000 ± 0.0000
      dataflow_match_score: 0.0000 ± 0.0000
    average_best_levenshtein: 0.0000 ± 0.0000
    average_best_normalized_levenshtein: 0.0000 ± 0.0000
    average_execution_time: 6.0124 ± 5.3631
  complex:
    count: 1.0000 ± 0.0000
    full_matches: 1.0000 ± 0.0000
    pv_exact_matched: 1.0000 ± 0.0000
    pv_exact_mismatched: 0.0000 ± 0.0000
    pv_exact_match_rate: 100.0000 ± 0.0000
    average_pv_match_rate: 100.0000 ± 0.0000
    average_pv_mismatch_rate: 0.0000 ± 0.0000
    timing_matched: 1.0000 ± 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: 1.0000 ± 0.0000
    full_mismatched: 0.0000 ± 0.0000
    full_match_rate: 100.0000 ± 0.0000
    accuracy: 100.0000 ± 0.0000
    average_best_codebleu:
      codebleu: 0.6951 ± 0.0000
      ngram_match_score: 0.3826 ± 0.0000
      weighted_ngram_match_score: 0.3976 ± 0.0000
      syntax_match_score: 1.0000 ± 0.0000
      dataflow_match_score: 1.0000 ± 0.0000
    average_best_levenshtein: 2.0000 ± 0.0000
    average_best_normalized_levenshtein: 0.0282 ± 0.0000
    average_execution_time: 4.4939 ± 1.5809
==================================================
