LLM performance on harder problems versus prerequisite skills

According to a user’s summary of the paper, humans scored 79.6% overall, and 72.7% of their correct answers also had every tested prerequisite correct. QWEN3-80B-INSTRUCT scored higher at 92.5%, but met that prerequisite standard for only 48.16% of its correct answers. The user argues that high accuracy can hide disconnected pockets of knowledge and says the paper recommends evaluating reasoning models with connected sets of easy and hard problems, not isolated benchmark questions alone.