Generating Data Driven Reasoning Rubrics

Build granular error taxonomies from incorrect reasoning traces, then use those rubrics to detect errors in LLM outputs across technical domains. Use when asked to: 'build a rubric for evaluating code solutions', 'create an error taxonomy for math reasoning', 'grade reasoning traces for correctness', 'build a reward function for domain-specific tasks', 'classify errors in chain-of-thought outputs', 'evaluate LLM reasoning without gold labels'.

ndpvt-web Updated 14 repo stars

File contents

ndpvt-web/arxiv-claude-skills commit ab4f390aa7

Frequently asked questions

npx get-skill add ndpvt-web/generating-data-driven-reasoning-rubrics