The Method / Architecture Synthesis course

Architecture Synthesis — Course

The whole derivation method as a sequence of lessons. Each pairs the book’s prose with a short exercise you run on a system you own; progress is tracked locally in your browser, no account required.

Part 0 — The problem

1 lesson

1Two Teams, Neither Wrong
Two teams built the same ticketing product with architectures that share nothing — and neither was wrong, because both derived correctly from different answers.

Part I — The toolkit

4 lessons

1The Answer Sheet
Nine questions, five driver modes, and an entry gate that prices wishes before they're allowed to count as answers.
2The Six Axes and the Ledger
Six axes, each value priced as provides / mechanism / always-on cost — the middle space between demands and named architecture styles.
3The Derivation
`next_step`: the procedure that turns a priced answer sheet and a ledger into a vector — null vector, prune, press, resolve, verify.
4Verification
The exit gate: the consistency lens and five rules of budget arithmetic that catch a wrong vector before anything is built.

Part II — Proof

3 lessons

1Three Profiles, One Domain
One domain, three answer sheets, three forced vectors — ticketing re-derived at venue, regional, and enterprise scale with every step citing its rule.
2Systems Nobody Asked Us to Derive
Four real systems — Stack Overflow, Shopify, Discord, a UK government registrar — derived blind from public commitments, with predictions registered before the outcomes were checked.
3When the Derivation Says No
Three reasonable answers on one trading platform empty an entire axis — the derivation halts, mechanically, and hands back a priced menu instead of an architecture.

Part III — The derivative

3 lessons

1Architecture as a Derivative
`next_step` run against a living system, not a blank page: the audit that flags unforced positions, incremental recomputation, and merge as a first-class output.
2Migration as Pathfinding
Migration as pathfinding: deltas that don't commute, intermediate states with physics of their own, and six qualitative indicators that weigh a route.
3Brownfield
Universal Credit, the UK's welfare platform: the method run against a real inherited system, audited by a decade of public record rather than by its author.

Part IV — What remains

2 lessons

1The Judgment That Stays Human
The inventory of decisions the method hands back to humans — and the MTTR/MTBF debate dissolved as a worked example of where mechanization stops.
2Closing
The derivative reframe, restated once more, and the book's standing invitation: run the worksheet on something real and send back what breaks it.