A Fusion run moves through five stages: Discovery, Plan, Build, Review, and Output. Each stage creates a different artifact, owns a different decision, and hands a narrower claim to the next stage.
Discovery turns an idea into intent
Discovery is divergent and ungated. It explores the user, problem, desired outcome, constraints, and unresolved choices. Its output is an Intent Brief, not code.
The useful question is: what must be true when this work is complete?
Plan freezes the proof contract
Planning converts intent into a versioned RunSpec with a Definition of Good, scope, non-goals, baseline, regression guards, risk surface, and gate command.
An independent plan screen can pass, revise, or block the contract. Gate authoring then binds protected checks to base and head coordinates.
Build explores and synthesizes
A solo profile runs one builder. A tournament runs candidates, an objective gate, blind judges, competing integrators, and a second jury pass.
The stage ends with one selected final, not with a remote mutation.
Review challenges the selected result
Fresh QA review catches immediate ship blockers. A separate deep audit tests production readiness, regression risk, security, and evidence completeness.
A failed review routes to bounded repair and re-verification. A passing review reaches ready to ship.
Output records the decision and returns authority
The Receipt links contract, subject, evidence, provenance, limitations, and delivery state. A human decides whether to ship. Closeout runs after durable delivery evidence exists.
The five-stage structure keeps creative work and proof work connected without confusing them. Every handoff asks the next stage to establish a smaller, more defensible fact.