Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 85.7143 ± 0.0000
total_entries: 126.0000 ± 0.0000
correct_matches: 108.0000 ± 0.0000
average_execution_time: 1.0492 ± 0.1467
metrics_by_complexity:
  simple:
    count: 120.0000 ± 0.0000
    full_matches: 105.0000 ± 0.0000
    exact_code_matches: 100.0000 ± 0.0000
    exact_code_match_rate: 83.3333 ± 0.0000
    pv_exact_matched: 105.0000 ± 0.0000
    pv_exact_mismatched: 15.0000 ± 0.0000
    pv_exact_match_rate: 87.5000 ± 0.0000
    average_pv_match_rate: 93.3333 ± 0.0000
    average_pv_mismatch_rate: 6.6667 ± 0.0000
    timing_matched: 119.0000 ± 0.0000
    timing_mismatched: 1.0000 ± 0.0000
    timing_match_rate: 99.1667 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 105.0000 ± 0.0000
    full_mismatched: 15.0000 ± 0.0000
    full_match_rate: 87.5000 ± 0.0000
    accuracy: 87.5000 ± 0.0000
    average_timing_score: 0.9917 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9450 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5792 ± 0.0000
      comb_7_CodeBLEU: 0.8230 ± 0.0000
      dataflow_match_score: 0.1333 ± 0.0000
      ngram_match_score: 0.2190 ± 0.0000
      syntax_match_score: 0.9711 ± 0.0000
      weighted_ngram_match_score: 0.1269 ± 0.0000
    average_best_levenshtein: 2.9833 ± 0.0000
    average_best_normalized_levenshtein: 0.0582 ± 0.0000
    average_inference_time: 0.9748 ± 0.0911
  complex:
    count: 6.0000 ± 0.0000
    full_matches: 3.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 16.6667 ± 0.0000
    pv_exact_matched: 3.0000 ± 0.0000
    pv_exact_mismatched: 3.0000 ± 0.0000
    pv_exact_match_rate: 50.0000 ± 0.0000
    average_pv_match_rate: 86.7210 ± 0.0000
    average_pv_mismatch_rate: 13.2790 ± 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: 3.0000 ± 0.0000
    full_mismatched: 3.0000 ± 0.0000
    full_match_rate: 50.0000 ± 0.0000
    accuracy: 50.0000 ± 0.0000
    average_timing_score: 0.9206 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.8779 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.6642 ± 0.0000
      comb_7_CodeBLEU: 0.7187 ± 0.0000
      dataflow_match_score: 0.7220 ± 0.0000
      ngram_match_score: 0.5000 ± 0.0000
      syntax_match_score: 0.7879 ± 0.0000
      weighted_ngram_match_score: 0.6470 ± 0.0000
    average_best_levenshtein: 40.0000 ± 0.0000
    average_best_normalized_levenshtein: 0.2198 ± 0.0000
    average_inference_time: 2.5360 ± 1.4959
==================================================
