Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 88.4804 ± 1.1232
total_entries: 136.0000 ± 0.0000
correct_matches: 120.3333 ± 1.5275
average_execution_time: 6.4398 ± 0.3531
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 111.3333 ± 0.5774
    exact_code_matches: 104.0000 ± 2.0000
    exact_code_match_rate: 89.6552 ± 1.7241
    pv_exact_matched: 114.6667 ± 0.5774
    pv_exact_mismatched: 1.3333 ± 0.5774
    pv_exact_match_rate: 98.8506 ± 0.4977
    average_pv_match_rate: 99.4253 ± 0.4977
    average_pv_mismatch_rate: 0.5747 ± 0.4977
    timing_matched: 111.3333 ± 0.5774
    timing_mismatched: 4.6667 ± 0.5774
    timing_match_rate: 95.9770 ± 0.4977
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 111.3333 ± 0.5774
    full_mismatched: 4.6667 ± 0.5774
    full_match_rate: 95.9770 ± 0.4977
    accuracy: 95.9770 ± 0.4977
    average_timing_score: 0.9802 ± 0.0044
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9914 ± 0.0048
    average_best_codebleu:
      average_CodeBLEU: 0.6203 ± 0.0049
      comb_7_CodeBLEU: 0.8467 ± 0.0021
      dataflow_match_score: 0.2730 ± 0.0050
      ngram_match_score: 0.2847 ± 0.0118
      syntax_match_score: 0.9952 ± 0.0005
      weighted_ngram_match_score: 0.2015 ± 0.0074
    average_best_levenshtein: 2.0833 ± 0.8539
    average_best_normalized_levenshtein: 0.0251 ± 0.0050
    average_inference_time: 4.8513 ± 0.3657
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 9.0000 ± 1.7321
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 9.6667 ± 1.5275
    pv_exact_mismatched: 10.3333 ± 1.5275
    pv_exact_match_rate: 48.3333 ± 7.6376
    average_pv_match_rate: 84.4687 ± 2.7633
    average_pv_mismatch_rate: 15.5313 ± 2.7633
    timing_matched: 15.6667 ± 1.1547
    timing_mismatched: 4.3333 ± 1.1547
    timing_match_rate: 78.3333 ± 5.7735
    temp_matched: 8.3333 ± 1.1547
    temp_mismatched: 0.6667 ± 1.1547
    temp_match_rate: 92.5926 ± 12.8300
    full_matched: 9.0000 ± 1.7321
    full_mismatched: 11.0000 ± 1.7321
    full_match_rate: 45.0000 ± 8.6603
    accuracy: 45.0000 ± 8.6603
    average_timing_score: 0.8061 ± 0.0266
    average_temp_score: 0.9874 ± 0.0150
    average_full_score: 0.8471 ± 0.0160
    average_best_codebleu:
      average_CodeBLEU: 0.4334 ± 0.0182
      comb_7_CodeBLEU: 0.5672 ± 0.0106
      dataflow_match_score: 0.5379 ± 0.0115
      ngram_match_score: 0.1970 ± 0.0318
      syntax_match_score: 0.6582 ± 0.0060
      weighted_ngram_match_score: 0.2239 ± 0.0304
    average_best_levenshtein: 149.1333 ± 6.4781
    average_best_normalized_levenshtein: 0.4291 ± 0.0153
    average_inference_time: 15.6529 ± 0.3791
==================================================
