Product Concept

Study before inventing, then diverge for real. Patterns exist to seed divergence — not to copy, but to know which conventions carry user understanding worth preserving and which persist because everyone copies everyone; the best raw material is often outside the category entirely (claims review owes more to GitHub code review or Stripe reconciliation than to competing estimating tools). Then the divergence itself, where LLMs fail catastrophically: cards, cards-with-sidebar, cards-with-tabs is one idea in three outfits. Concepts are different answers to how should this problem fundamentally work. Divergence is cheap here and unaffordable after code exists — so storm wide first, then converge.

Part of the product-define suite. Read design/problem.md (Desire, sandbox card, aging) and model.md; write design/patterns.md and design/concepts.md; update context.md's Current concept after selection. Gate 1 must pass first. Low-uncertainty routes may run the patterns pass alone. Full diverge: patterns → storm → shortlist → specs → Browse → Cut 1 → Compare → Cut 2 → Commit → subtract.

Quick Reference

Core Principles