One governed learning loop
Goal → Context → Rules → Blueprint → Jobs → Tests → Feedback
This is not a one-way conveyor from a request to generated code. Each stage answers a different decision, and Feedback can return evidence to whichever upstream stage must change.
01. Goal: Start with the outcome.
Define the desired change, accountable owner, reason, constraints, and measurable direction for success.
02. Context: Bring the system into view.
Assemble the relevant systems, code, data, documents, decisions, dependencies, people, and domain knowledge.
03. Rules: Agree on what must be true.
Make the approved conditions for success explicit, authoritative downstream, and reviewable when they change.
04. Blueprint: Design before generation.
Define boundaries, interfaces, data flows, security, migration logic, risks, and the verification approach before Jobs begin.
05. Jobs: The Blueprint becomes executable.
Turn the approved design into bounded, permissioned, and attributable units of execution.
06. Tests: Tested before trusted.
Produce evidence that the exact change follows the approved Rules and Blueprint before it is treated as complete.
07. Feedback: The outcome becomes the next input.
Return production results, user feedback, incidents, and business evidence to the appropriate upstream stage.
Human authority remains visible
A gate is a real decision. Accountable owners approve scope and design, reviewers evaluate the required evidence, and production learning is routed back into the system rather than automatically becoming new work.
Continue with the Goal guide to start the lifecycle from an outcome rather than an implementation request.