Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 90.1961 ± 0.4245
total_entries: 136.0000 ± 0.0000
correct_matches: 122.6667 ± 0.5774
average_execution_time: 5.1851 ± 0.2271
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 111.6667 ± 0.5774
    exact_code_matches: 107.3333 ± 1.5275
    exact_code_match_rate: 92.5287 ± 1.3168
    pv_exact_matched: 112.0000 ± 0.0000
    pv_exact_mismatched: 4.0000 ± 0.0000
    pv_exact_match_rate: 96.5517 ± 0.0000
    average_pv_match_rate: 97.5862 ± 0.3982
    average_pv_mismatch_rate: 2.4138 ± 0.3982
    timing_matched: 112.3333 ± 0.5774
    timing_mismatched: 3.6667 ± 0.5774
    timing_match_rate: 96.8391 ± 0.4977
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 111.6667 ± 0.5774
    full_mismatched: 4.3333 ± 0.5774
    full_match_rate: 96.2644 ± 0.4977
    accuracy: 96.2644 ± 0.4977
    average_timing_score: 0.9738 ± 0.0043
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9755 ± 0.0040
    average_best_codebleu:
      average_CodeBLEU: 0.6023 ± 0.0019
      comb_7_CodeBLEU: 0.8356 ± 0.0008
      dataflow_match_score: 0.2414 ± 0.0086
      ngram_match_score: 0.2615 ± 0.0054
      syntax_match_score: 0.9824 ± 0.0027
      weighted_ngram_match_score: 0.1652 ± 0.0037
    average_best_levenshtein: 1.0575 ± 0.0782
    average_best_normalized_levenshtein: 0.0270 ± 0.0026
    average_inference_time: 4.4522 ± 0.2557
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 11.0000 ± 1.0000
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 13.3333 ± 1.5275
    pv_exact_mismatched: 6.6667 ± 1.5275
    pv_exact_match_rate: 66.6667 ± 7.6376
    average_pv_match_rate: 93.7051 ± 2.2358
    average_pv_mismatch_rate: 6.2949 ± 2.2358
    timing_matched: 15.0000 ± 0.0000
    timing_mismatched: 5.0000 ± 0.0000
    timing_match_rate: 75.0000 ± 0.0000
    temp_matched: 7.6667 ± 0.5774
    temp_mismatched: 1.3333 ± 0.5774
    temp_match_rate: 85.1852 ± 6.4150
    full_matched: 11.0000 ± 1.0000
    full_mismatched: 9.0000 ± 1.0000
    full_match_rate: 55.0000 ± 5.0000
    accuracy: 55.0000 ± 5.0000
    average_timing_score: 0.8506 ± 0.0136
    average_temp_score: 0.9700 ± 0.0213
    average_full_score: 0.9188 ± 0.0153
    average_best_codebleu:
      average_CodeBLEU: 0.4376 ± 0.0200
      comb_7_CodeBLEU: 0.5551 ± 0.0113
      dataflow_match_score: 0.5398 ± 0.0395
      ngram_match_score: 0.2243 ± 0.0414
      syntax_match_score: 0.6769 ± 0.0161
      weighted_ngram_match_score: 0.2594 ± 0.0329
    average_best_levenshtein: 137.1000 ± 15.2487
    average_best_normalized_levenshtein: 0.4197 ± 0.0334
    average_inference_time: 9.4360 ± 0.6643
==================================================
