"author:razilzul" matching MCP servers:

Enables coding agents to evaluate a single model over a golden set with trap cases, reporting accuracy overall and on traps, mean latency, cost per request, and failing cases. Also compares a baseline against a candidate model and returns a pass/fail gate verdict with failure reasons, so a model swap can be checked before it ships.