EvalPlus / MBPP+ C2 summary
============================================================

claude-sonnet-4-6 / readability
  n = 224
  original MBPP pass = 178 (79.464%)
  EvalPlus base pass = 178 (79.464%)
  EvalPlus plus pass = 151 (67.411%)
  original PASS -> base FAIL = 13
  original FAIL -> base PASS = 13
  total base/original mismatches = 26 (11.607%)

claude-sonnet-4-6 / standard
  n = 224
  original MBPP pass = 184 (82.143%)
  EvalPlus base pass = 181 (80.804%)
  EvalPlus plus pass = 157 (70.089%)
  original PASS -> base FAIL = 15
  original FAIL -> base PASS = 12
  total base/original mismatches = 27 (12.054%)

gemini-2.5-pro / readability
  n = 224
  original MBPP pass = 162 (72.321%)
  EvalPlus base pass = 161 (71.875%)
  EvalPlus plus pass = 138 (61.607%)
  original PASS -> base FAIL = 12
  original FAIL -> base PASS = 11
  total base/original mismatches = 23 (10.268%)

gemini-2.5-pro / standard
  n = 224
  original MBPP pass = 161 (71.875%)
  EvalPlus base pass = 159 (70.982%)
  EvalPlus plus pass = 137 (61.161%)
  original PASS -> base FAIL = 12
  original FAIL -> base PASS = 10
  total base/original mismatches = 22 (9.821%)

gpt-5.4 / readability
  n = 224
  original MBPP pass = 153 (68.304%)
  EvalPlus base pass = 152 (67.857%)
  EvalPlus plus pass = 134 (59.821%)
  original PASS -> base FAIL = 11
  original FAIL -> base PASS = 10
  total base/original mismatches = 21 (9.375%)

gpt-5.4 / standard
  n = 224
  original MBPP pass = 156 (69.643%)
  EvalPlus base pass = 154 (68.750%)
  EvalPlus plus pass = 138 (61.607%)
  original PASS -> base FAIL = 11
  original FAIL -> base PASS = 9
  total base/original mismatches = 20 (8.929%)

Overall
  n = 1344
  original MBPP pass = 994 (73.958%)
  EvalPlus base pass = 985 (73.289%)
  EvalPlus plus pass = 855 (63.616%)
  base/original mismatches = 139 (10.342%)
