exploring-an-idea

List every claim the idea rests on and label each as measured, documented, or assumed. Any documented or assumed claim whose failure would change the design becomes a probe task: a cheap script that answers one question with a number before the document is written.

A documented claim comes from vendor documentation and is not verification; a measured claim comes from running a probe and seeing the result with a cited number. The Skill treats documented load-bearing claims as tasks, not facts, because vendor docs are a common source of false premises.

Yes, but in Claude Desktop you can only do research, not run probes. The Skill names each probe, states what number would settle it, and hands the probes over as the first work item; Claude Code can write and run the probes directly.

Run the saturation and trademark pass before the document is titled and before the repo is created, since renaming gets more expensive with every commit. The repo name follows the name check, and domain and handle availability are checked while the name is still free.

Constructed samples hide distributional facts that change the answer. In the cited case, a real collection was 28% 7-inch singles that resolve at 0%, so a flat rate across a constructed sample would have answered the wrong question entirely.

A Phase 0 spike only helps if its exit criteria are numbers rather than activities like outreach or investigation. Also, running lacquer init inside a throwaway spike records the spike's stack as the project's stack, so the full stack should be declared up front with --stack.