product-validate
Product Validate
One loop, two motions: attack the riskiest assumption, then judge what happened. Prototype the risk, not the prettiest screen — often the risk isn't UI at all but whether the data can be found, whether classification is reliable, whether users understand the words. Then evaluation that goes far beyond "did users click the button": nine lenses, each able to fail a design that passes the other eight, with findings routed backward — a comprehension failure to IA or framing, a trust failure to the model — never patched with a tooltip at the surface.
Part of the product-define suite. Read design/concepts.md (risks), flows.md, model.md (⚠️ flags and error ledger), problem.md (signals, non-goals); write design/validation.md; update context.md. Gate 4 runs against this file.
Quick Reference
| Topic | File | Read when |
|---|---|---|
| Assumption → cheapest instrument | assets/instruments.md | Before building a prototype |
| Nine lenses + routing | assets/nine-lenses.md | Every evaluation |
Core Principles
-
Name the riskiest assumption first — then pick an instrument from assets/instruments.md. Band-aware: Stretch/Wild need explicit plans.
-
Match instrument to question — table in instruments.md.