Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 83.0882 ± 0.0000
total_entries: 136.0000 ± 0.0000
correct_matches: 113.0000 ± 0.0000
average_execution_time: 1.1356 ± 0.0251
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 107.0000 ± 0.0000
    exact_code_matches: 88.3333 ± 0.5774
    exact_code_match_rate: 76.1494 ± 0.4977
    pv_exact_matched: 110.0000 ± 0.0000
    pv_exact_mismatched: 6.0000 ± 0.0000
    pv_exact_match_rate: 94.8276 ± 0.0000
    average_pv_match_rate: 96.2644 ± 0.0000
    average_pv_mismatch_rate: 3.7356 ± 0.0000
    timing_matched: 112.0000 ± 0.0000
    timing_mismatched: 4.0000 ± 0.0000
    timing_match_rate: 96.5517 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 107.0000 ± 0.0000
    full_mismatched: 9.0000 ± 0.0000
    full_match_rate: 92.2414 ± 0.0000
    accuracy: 92.2414 ± 0.0000
    average_timing_score: 0.9819 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9665 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5576 ± 0.0008
      comb_7_CodeBLEU: 0.7989 ± 0.0015
      dataflow_match_score: 0.2241 ± 0.0000
      ngram_match_score: 0.1961 ± 0.0009
      syntax_match_score: 0.9195 ± 0.0041
      weighted_ngram_match_score: 0.1148 ± 0.0002
    average_best_levenshtein: 1.3305 ± 0.0348
    average_best_normalized_levenshtein: 0.0555 ± 0.0009
    average_inference_time: 1.0789 ± 0.1154
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 6.0000 ± 0.0000
    exact_code_matches: 1.0000 ± 0.0000
    exact_code_match_rate: 5.0000 ± 0.0000
    pv_exact_matched: 7.0000 ± 0.0000
    pv_exact_mismatched: 13.0000 ± 0.0000
    pv_exact_match_rate: 35.0000 ± 0.0000
    average_pv_match_rate: 53.8302 ± 1.0825
    average_pv_mismatch_rate: 46.1698 ± 1.0825
    timing_matched: 14.6667 ± 0.5774
    timing_mismatched: 5.3333 ± 0.5774
    timing_match_rate: 73.3333 ± 2.8868
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 9.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 6.0000 ± 0.0000
    full_mismatched: 14.0000 ± 0.0000
    full_match_rate: 30.0000 ± 0.0000
    accuracy: 30.0000 ± 0.0000
    average_timing_score: 0.8154 ± 0.0091
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.5700 ± 0.0048
    average_best_codebleu:
      average_CodeBLEU: 0.3767 ± 0.0051
      comb_7_CodeBLEU: 0.4893 ± 0.0055
      dataflow_match_score: 0.5377 ± 0.0086
      ngram_match_score: 0.1698 ± 0.0038
      syntax_match_score: 0.5909 ± 0.0030
      weighted_ngram_match_score: 0.2085 ± 0.0051
    average_best_levenshtein: 138.1167 ± 2.5106
    average_best_normalized_levenshtein: 0.4461 ± 0.0006
    average_inference_time: 1.4647 ± 0.6049
==================================================
