Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 88.2353 ± 0.7353
total_entries: 136.0000 ± 0.0000
correct_matches: 120.0000 ± 1.0000
average_execution_time: 0.8244 ± 0.0864
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 111.3333 ± 0.5774
    exact_code_matches: 104.3333 ± 0.5774
    exact_code_match_rate: 89.9425 ± 0.4977
    pv_exact_matched: 114.0000 ± 0.0000
    pv_exact_mismatched: 2.0000 ± 0.0000
    pv_exact_match_rate: 98.2759 ± 0.0000
    average_pv_match_rate: 99.4923 ± 0.0830
    average_pv_mismatch_rate: 0.5077 ± 0.0830
    timing_matched: 112.6667 ± 0.5774
    timing_mismatched: 3.3333 ± 0.5774
    timing_match_rate: 97.1264 ± 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: 0.9877 ± 0.0024
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9935 ± 0.0005
    average_best_codebleu:
      average_CodeBLEU: 0.6267 ± 0.0019
      comb_7_CodeBLEU: 0.8488 ± 0.0012
      dataflow_match_score: 0.2845 ± 0.0000
      ngram_match_score: 0.2991 ± 0.0031
      syntax_match_score: 0.9939 ± 0.0015
      weighted_ngram_match_score: 0.2137 ± 0.0031
    average_best_levenshtein: 1.3678 ± 0.0407
    average_best_normalized_levenshtein: 0.0221 ± 0.0017
    average_inference_time: 0.7072 ± 0.1046
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 8.6667 ± 0.5774
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 10.6667 ± 0.5774
    pv_exact_mismatched: 9.3333 ± 0.5774
    pv_exact_match_rate: 53.3333 ± 2.8868
    average_pv_match_rate: 84.8739 ± 1.3395
    average_pv_mismatch_rate: 15.1261 ± 1.3395
    timing_matched: 13.3333 ± 0.5774
    timing_mismatched: 6.6667 ± 0.5774
    timing_match_rate: 66.6667 ± 2.8868
    temp_matched: 6.0000 ± 0.0000
    temp_mismatched: 3.0000 ± 0.0000
    temp_match_rate: 66.6667 ± 0.0000
    full_matched: 8.6667 ± 0.5774
    full_mismatched: 11.3333 ± 0.5774
    full_match_rate: 43.3333 ± 2.8868
    accuracy: 43.3333 ± 2.8868
    average_timing_score: 0.7775 ± 0.0125
    average_temp_score: 0.9372 ± 0.0001
    average_full_score: 0.8303 ± 0.0091
    average_best_codebleu:
      average_CodeBLEU: 0.4120 ± 0.0045
      comb_7_CodeBLEU: 0.5217 ± 0.0021
      dataflow_match_score: 0.4343 ± 0.0054
      ngram_match_score: 0.2189 ± 0.0110
      syntax_match_score: 0.6554 ± 0.0005
      weighted_ngram_match_score: 0.2395 ± 0.0094
    average_best_levenshtein: 140.3833 ± 2.6383
    average_best_normalized_levenshtein: 0.4486 ± 0.0093
    average_inference_time: 1.5045 ± 0.0388
==================================================
