ZIm/crates/eval/examples/optimizer_schema_refactor/prompt.md

I need to refactor our codebase to improve the clarity and consistency of our data model, particularly around how we handle labeled evaluation data for our threshold optimization system. Currently, the naming and structure might imply that this data is only used for testing, when in reality it represents labeled examples that power both training and evaluation. The changes should better reflect that these are curated data points with known outcomes, not just test cases. Focus on updating the core data model and ensuring all dependent components—like the cache optimizer, router, and evaluation utilities—properly reference this updated concept. The implementation should maintain all existing functionality while making the naming more semantically accurate. Where relevant, consider adding parameters to fine-tune optimization behavior, like allowing control over the granularity of threshold searches.