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: 0.8509 ± 0.1227
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 98.0000 ± 0.0000
    exact_code_matches: 93.0000 ± 0.0000
    exact_code_match_rate: 77.5000 ± 0.0000
    pv_exact_matched: 98.0000 ± 0.0000
    pv_exact_mismatched: 22.0000 ± 0.0000
    pv_exact_match_rate: 81.6667 ± 0.0000
    average_pv_match_rate: 83.7500 ± 0.0000
    average_pv_mismatch_rate: 16.2500 ± 0.0000
    timing_matched: 103.0000 ± 0.0000
    timing_mismatched: 17.0000 ± 0.0000
    timing_match_rate: 85.8333 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 98.0000 ± 0.0000
    full_mismatched: 22.0000 ± 0.0000
    full_match_rate: 81.6667 ± 0.0000
    accuracy: 81.6667 ± 0.0000
    average_timing_score: 0.8573 ± 0.0000
    average_temp_score: 0.9500 ± 0.0000
    average_full_score: 0.8415 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5301 ± 0.0000
      comb_7_CodeBLEU: 0.7751 ± 0.0000
      dataflow_match_score: 0.0917 ± 0.0000
      ngram_match_score: 0.1698 ± 0.0000
      syntax_match_score: 0.8767 ± 0.0000
      weighted_ngram_match_score: 0.0740 ± 0.0000
    average_best_levenshtein: 3.5000 ± 0.0000
    average_best_normalized_levenshtein: 0.1012 ± 0.0000
    average_inference_time: 0.7598 ± 0.0656
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 5.0000 ± 0.0000
    exact_code_matches: 2.0000 ± 0.0000
    exact_code_match_rate: 33.3333 ± 0.0000
    pv_exact_matched: 5.0000 ± 0.0000
    pv_exact_mismatched: 1.0000 ± 0.0000
    pv_exact_match_rate: 83.3333 ± 0.0000
    average_pv_match_rate: 91.6667 ± 0.0000
    average_pv_mismatch_rate: 8.3333 ± 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: 5.0000 ± 0.0000
    full_mismatched: 1.0000 ± 0.0000
    full_match_rate: 83.3333 ± 0.0000
    accuracy: 83.3333 ± 0.0000
    average_timing_score: 0.9988 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9331 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.7944 ± 0.0000
      comb_7_CodeBLEU: 0.8847 ± 0.0000
      dataflow_match_score: 0.9359 ± 0.0000
      ngram_match_score: 0.6291 ± 0.0000
      syntax_match_score: 0.9540 ± 0.0000
      weighted_ngram_match_score: 0.6584 ± 0.0000
    average_best_levenshtein: 14.0000 ± 0.0000
    average_best_normalized_levenshtein: 0.0713 ± 0.0000
    average_inference_time: 2.6738 ± 1.2646
==================================================
