O1 · Specialist consultation
O1 contains 103 questions from 42 cases. questions.jsonl provides case information and consultation questions; rubrics.jsonl provides 1,726 weighted criteria.
The 1,069 positive criteria cover core conclusions, supporting evidence and conditional recommendations. The 657 negative criteria cover 337 major and 320 minor errors. A true positive criterion earns credit; a true negative criterion incurs its specified deduction.
Generate one answer per question, then judge each criterion against that answer. The paper uses GPT-OSS-120B as its default judge. Scores average questions equally; case-weighted results are reported separately.