Prompt Evaluations

The evaluation system uses a Prompt Quality Rubric with eight dimensions:

1. Objective clarity.
2. Context sufficiency.
3. Instruction completeness.
4. Constraint clarity.
5. Output contract quality.
6. Grounding and uncertainty handling.
7. Eval readiness.
8. Efficiency and safety.

The examples in catalog/evaluations.json show before-and-after prompt changes for support routing, RAG policy answers, and coding-agent repository changes. Each example records before scores, after scores, related patterns, and acceptance criteria.

Example files are also available in examples/evaluations/.