- replace original/optimized semantics with analysis, result, and compare request types - centralize evaluation prompt construction in shared builders - tighten rendered sample and evidence behavior tests for core evaluation service