I've started to assemble the parts to build an experimental benchmark setup that I can use to develop the purpose driven ideas. It's a new public/open source GitHub organization https://github.com/brazil-bench that contains a repeatable development task "make an MCP server that knows about Brazilian Football" and repo's for my first two manually operated attempts at this work, as well as the beginnings of a pourpoise framework to automate running new versions of this test. If anyone wants to collaborate on this work, let me know.