Glossary term
Glossary term
Evaluation and Benchmarks
Abbreviation for unsupported-claim rate.
Created for this library
An LLM safety team monitors UCR to confirm the model is not generating unsupported claims in factual contexts.
A factuality evaluation team reports UCR in its model card so downstream users see the rate of unsupported claims.
A research lab tracks UCR across model versions as a leading indicator of grounding quality.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License