Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 86.0294 ± 0.0000
total_entries: 136.0000 ± 0.0000
correct_matches: 117.0000 ± 0.0000
average_execution_time: 27.2206 ± 0.2672
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 111.0000 ± 0.0000
    exact_code_matches: 103.0000 ± 0.0000
    exact_code_match_rate: 88.7931 ± 0.0000
    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: 98.9943 ± 0.0000
    average_pv_mismatch_rate: 1.0057 ± 0.0000
    timing_matched: 113.0000 ± 0.0000
    timing_mismatched: 3.0000 ± 0.0000
    timing_match_rate: 97.4138 ± 0.0000
    temp_matched: 0.0000 ± 0.0000
    temp_mismatched: 0.0000 ± 0.0000
    temp_match_rate: 0.0000 ± 0.0000
    full_matched: 111.0000 ± 0.0000
    full_mismatched: 5.0000 ± 0.0000
    full_match_rate: 95.6897 ± 0.0000
    accuracy: 95.6897 ± 0.0000
    average_timing_score: 0.9896 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9899 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.6011 ± 0.0000
      comb_7_CodeBLEU: 0.8320 ± 0.0000
      dataflow_match_score: 0.2931 ± 0.0000
      ngram_match_score: 0.2613 ± 0.0000
      syntax_match_score: 0.9719 ± 0.0000
      weighted_ngram_match_score: 0.1713 ± 0.0000
    average_best_levenshtein: 1.5690 ± 0.0000
    average_best_normalized_levenshtein: 0.0340 ± 0.0000
    average_inference_time: 8.2511 ± 0.0545
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 6.0000 ± 0.0000
    exact_code_matches: 2.0000 ± 0.0000
    exact_code_match_rate: 10.0000 ± 0.0000
    pv_exact_matched: 6.0000 ± 0.0000
    pv_exact_mismatched: 14.0000 ± 0.0000
    pv_exact_match_rate: 30.0000 ± 0.0000
    average_pv_match_rate: 41.4596 ± 0.3972
    average_pv_mismatch_rate: 58.5404 ± 0.3972
    timing_matched: 10.3333 ± 0.5774
    timing_mismatched: 9.6667 ± 0.5774
    timing_match_rate: 51.6667 ± 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: 0.5804 ± 0.0260
    average_temp_score: 0.5500 ± 0.0000
    average_full_score: 0.4389 ± 0.0072
    average_best_codebleu:
      average_CodeBLEU: 0.4052 ± 0.0070
      comb_7_CodeBLEU: 0.5327 ± 0.0060
      dataflow_match_score: 0.5545 ± 0.0026
      ngram_match_score: 0.1594 ± 0.0034
      syntax_match_score: 0.5811 ± 0.0127
      weighted_ngram_match_score: 0.2258 ± 0.0140
    average_best_levenshtein: 567.5667 ± 88.6233
    average_best_normalized_levenshtein: 0.5529 ± 0.0138
    average_inference_time: 137.2441 ± 2.0650
==================================================
