The make / keep grid is a menu. On its own it doesn't say where to start, what feeds what, or why the work should get cheaper over time. The order below does — because every step hands a finished output to the next, and the last step feeds the first.
Each arc ends by handing a named artifact to the next. That hand-off is the whole trick: you never start a step from a blank page, and you never re-derive what an earlier step already settled.
Derive the medium-specific design schema and tokens, then render against them.
Assemble the full deliverable bundle end-to-end when the output is a whole new thing.
Hold it to a craft constitution and an evidence-based Definition of Done before it ships.
Run adversarial, multi-lens review; return prioritized findings and one reusable lesson.
Capture that lesson as a durable asset and update the index — so the next cycle finds it and starts cheaper. This writes back into step 1.
The portfolio has run these four arcs for four cycles — not as a diagram, but literally. The proof is in the repo:
That’s the test of a sustainable order: you should be able to point at the artifacts each arc left behind. Watch the cycle run in the animation library, or start at the overview.