Statistics Summary for op agent (mean ± std):
==================================================
accuracy: 82.5980 ± 0.8490
total_entries: 136.0000 ± 0.0000
correct_matches: 112.3333 ± 1.1547
average_execution_time: 1.2697 ± 0.0883
metrics_by_complexity:
  simple:
    count: 116.0000 ± 0.0000
    full_matches: 107.0000 ± 0.0000
    exact_code_matches: 89.6667 ± 1.1547
    exact_code_match_rate: 77.2989 ± 0.9954
    pv_exact_matched: 111.0000 ± 0.0000
    pv_exact_mismatched: 5.0000 ± 0.0000
    pv_exact_match_rate: 95.6897 ± 0.0000
    average_pv_match_rate: 97.1264 ± 0.0000
    average_pv_mismatch_rate: 2.8736 ± 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: 0.9883 ± 0.0000
    average_temp_score: 1.0000 ± 0.0000
    average_full_score: 0.9747 ± 0.0000
    average_best_codebleu:
      average_CodeBLEU: 0.5738 ± 0.0006
      comb_7_CodeBLEU: 0.8064 ± 0.0002
      dataflow_match_score: 0.2241 ± 0.0000
      ngram_match_score: 0.2278 ± 0.0018
      syntax_match_score: 0.9228 ± 0.0000
      weighted_ngram_match_score: 0.1447 ± 0.0004
    average_best_levenshtein: 1.3621 ± 0.0299
    average_best_normalized_levenshtein: 0.0557 ± 0.0023
    average_inference_time: 1.1755 ± 0.0518
  complex:
    count: 20.0000 ± 0.0000
    full_matches: 5.3333 ± 1.1547
    exact_code_matches: 0.0000 ± 0.0000
    exact_code_match_rate: 0.0000 ± 0.0000
    pv_exact_matched: 5.3333 ± 1.1547
    pv_exact_mismatched: 14.6667 ± 1.1547
    pv_exact_match_rate: 26.6667 ± 5.7735
    average_pv_match_rate: 50.2958 ± 2.0805
    average_pv_mismatch_rate: 49.7042 ± 2.0805
    timing_matched: 15.0000 ± 1.0000
    timing_mismatched: 5.0000 ± 1.0000
    timing_match_rate: 75.0000 ± 5.0000
    temp_matched: 2.0000 ± 0.0000
    temp_mismatched: 7.0000 ± 0.0000
    temp_match_rate: 22.2222 ± 0.0000
    full_matched: 5.3333 ± 1.1547
    full_mismatched: 14.6667 ± 1.1547
    full_match_rate: 26.6667 ± 5.7735
    accuracy: 26.6667 ± 5.7735
    average_timing_score: 0.8126 ± 0.0193
    average_temp_score: 0.7455 ± 0.0234
    average_full_score: 0.5797 ± 0.0239
    average_best_codebleu:
      average_CodeBLEU: 0.3215 ± 0.0189
      comb_7_CodeBLEU: 0.4386 ± 0.0144
      dataflow_match_score: 0.4491 ± 0.0136
      ngram_match_score: 0.1136 ± 0.0204
      syntax_match_score: 0.5341 ± 0.0093
      weighted_ngram_match_score: 0.1392 ± 0.0322
    average_best_levenshtein: 151.7000 ± 14.8599
    average_best_normalized_levenshtein: 0.4831 ± 0.0250
    average_inference_time: 1.8164 ± 0.3132
==================================================
