ClarusC64/

ecb_v01

A dataset for frame-aware intelligence. Current LLMs often answer inside a broken question, reinforcing:

false assumptions invalid frames contradictions harmful premises

This benchmark evaluates the ability to stop, and clarify the premise before responding.

	Expected Actions




	

Action Meaning

clarify Identify instability in the prompt and restate what must be resolved first

There is no “answering” action in this… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/ecb_v01.