LLM evaluation metrics

The design and validation of scores used to judge model output, including composite metrics, their sub-components, and the failure modes that come from gating one measurement on another.