Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 41.6667 ± 0.4245
total_entries: 136.0000 ± 0.0000
correct_matches: 56.6667 ± 0.5774
average_execution_time: 2.1045 ± 0.0645
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 55.6667 ± 0.5774
    exact_code_matches: 46.0000 ± 0.0000
    exact_code_match_rate: 39.6552 ± 0.0000
    pv_exact_matched: 55.6667 ± 0.5774
    pv_exact_mismatched: 60.3333 ± 0.5774
    pv_exact_match_rate: 47.9885 ± 0.4977
    average_pv_match_rate: 54.7701 ± 0.4977
    average_pv_mismatch_rate: 45.2299 ± 0.4977
    timing_matched: 62.6667 ± 0.5774
    timing_mismatched: 53.3333 ± 0.5774
    timing_match_rate: 54.0230 ± 0.4977
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 6.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 55.6667 ± 0.5774
    full_mismatched: 60.3333 ± 0.5774
    full_match_rate: 47.9885 ± 0.4977
    accuracy: 47.9885 ± 0.4977
    average_timing_score: 0.5753 ± 0.0050
    average_temp_score: 0.9483 ± 0.0000
    average_full_score: 0.5470 ± 0.0050
    average_best_codebleu:
      average_CodeBLEU: 0.5305 ± 0.0000
      comb_7_CodeBLEU: 0.7696 ± 0.0000
      dataflow_match_score: 0.2845 ± 0.0000
      ngram_match_score: 0.1537 ± 0.0000
      syntax_match_score: 0.8580 ± 0.0000
      weighted_ngram_match_score: 0.1102 ± 0.0000
    average_best_levenshtein: 8.3966 ± 0.0000
    average_best_normalized_levenshtein: 0.2366 ± 0.0000
    average_inference_time: 1.9494 ± 0.0429
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 1.0000 ± 0.0000
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 1.0000 ± 0.0000
    pv_exact_mismatched: 19.0000 ± 0.0000
    pv_exact_match_rate: 5.0000 ± 0.0000
    average_pv_match_rate: 51.9401 ± 1.5221
    average_pv_mismatch_rate: 48.0599 ± 1.5221
    timing_matched: 8.0000 ± 0.0000
    timing_mismatched: 12.0000 ± 0.0000
    timing_match_rate: 40.0000 ± 0.0000
    temp_matched: 7.0000 ± 0.0000
    temp_mismatched: 2.0000 ± 0.0000
    temp_match_rate: 77.7778 ± 0.0000
    full_matched: 1.0000 ± 0.0000
    full_mismatched: 19.0000 ± 0.0000
    full_match_rate: 5.0000 ± 0.0000
    accuracy: 5.0000 ± 0.0000
    average_timing_score: 0.6881 ± 0.0110
    average_temp_score: 0.8992 ± 0.0000
    average_full_score: 0.5739 ± 0.0113
    average_best_codebleu:
      average_CodeBLEU: 0.3301 ± 0.0006
      comb_7_CodeBLEU: 0.4654 ± 0.0007
      dataflow_match_score: 0.4240 ± 0.0008
      ngram_match_score: 0.0945 ± 0.0010
      syntax_match_score: 0.5874 ± 0.0006
      weighted_ngram_match_score: 0.1145 ± 0.0003
    average_best_levenshtein: 163.2667 ± 1.7863
    average_best_normalized_levenshtein: 0.5112 ± 0.0044
    average_inference_time: 3.0043 ± 0.2045
==================================================
