A verify-and-improve loop for any task — code, prompt, doc, or design.
The harness is a verify-and-improve loop: it takes any task — building a feature, tuning a prompt, writing a doc — and first pins down a measurable "done-when", so finished is a fact you can check, not an opinion. Then it cycles: look at the real state, pick one bounded unit of work, do it, and prove it at a real boundary — never on a claim or a mock. It keeps cycling until the done-when is met, and at the end it doesn't just hand back the result: it delivers the work and publishes a companion course so the how is captured, not lost.