You start with an empty board. Draw the system this workload needs — the smallest one that can serve it — and press Go Live. Then traffic starts arriving, and it flows left to right through whatever you built: served requests earn money and bring more users, and more users break what worked yesterday.

A finished design is five to seven boxes. Only the ones this workload could use are on the board, and you will not need all of those. Everything you build bills from the moment it exists, so building it early and building it late are both mistakes — just different ones.

Every box shows how full it is, because latency is service time ÷ (1 − utilization) — systems do not slow down gradually, they fall off a cliff. Hover any box to interrogate it: traffic, latency, errors, saturation and run-rate. How much of that answer you get depends on how much telemetry you have bought.

Everything you own bills you every second in two parts: a fixed cost the moment it exists — the team who knows it, the dashboards, the 3am pages — plus a small amount per node. Your first node of anything is expensive; your tenth is cheap. Anything not carrying its weight can be scaled back down with −.

Stuck? The 🧘 Guru diagnoses the board live and names the next move. Management will also have opinions; 📣 mutes them.

When the run ends you get an architecture review — the one you would get at work, on what you built, when you built it, and what it cost. Those seven lines are also, not by accident, the seven things an interviewer is listening for. Space pauses.

It moves at the speed of a system that is growing, not at the speed of a game. Use 2× and 4× when nothing is happening, and ⏸ when it is.