reasoning-degeneration-dev/

t1-strategy-musr-baseline-kimi-k2-instruct-musr-base

Strategy compliance evaluation on MuSR murder mysteries — baseline variant. Model received standard MuSR cot+ prompt (baseline control). Judge still scores against evidence matrix rubric. Compliance is scored by an LLM judge (1-5 Likert) against the Evidence Matrix rubric.

	Results




	

Metric Value

pass@1 0.9000

Strategy compliance (mean) 3.50

Strategy compliance (min) 2

Strategy compliance (max) 5

Total… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/t1-strategy-musr-baseline-kimi-k2-instruct-musr-base.