MCQA critique: A) Forced B) Flawed C) Fixable D) All of the Above

"Which of These Best Describes Multiple Choice Evaluation with LLMs?" (ACL 2025): Balepur, Rudinger, and Boyd-Graber import educational-testing theory to argue the MCQA format itself — forced choice, answer leakage, no generation — systematically mismeasures LLM ability, and propose constructed-response and rubric-based fixes. A seed paper of the benchmark-science / philosophy-of-evaluation thread.