MMLU-Pro audit: longest-answer shortcut more than doubled

An audit of all 12,032 MMLU-Pro test questions found the "answer is D" position shortcut is gone, but picking the longest option now scores 18.38% versus 11.09% chance. New-source questions, 43% of the benchmark, carry 88.8% of the length excess, not the added distractors.

• Gold answer position in MMLU-Pro is uniform: chi-square 9.35 vs 27.88 threshold
• Pick-longest scores 18.38% against 11.09% chance
• Biology reaches 44.61% pick-longest versus 11.08% chance
• New questions are 43% of the benchmark but 88.8% of length excess