Standard agentic leaderboards measure models against shared, static tool schemas — which is a poor proxy for how a model handles the specific tools your stack actually exposes. This post argues for running evaluation against your own tooling and shows how to do it with open-weight models using the Transformers ecosystem. The practical upshot: models that look comparable on published benchmarks can diverge sharply when the tool signatures change. Worth running before committing to a model for an agentic workflow.