Pin a build-vs-research heuristic in CLAUDE.md

When a user repeatedly gives the same working-philosophy guidance, capture it as a short directive section in the project CLAUDE.md rather than leaving it in chat. A useful, generalizable rule to encode: build now if an idea is ~95% logically sound (coding is cheap), but if a claim lacks an order of magnitude of validation, do the experiment/math first — a wrong foundation is the costly mistake. Prefer no project CLAUDE.md exists check first before creating one.