Single choice
Pick the best fix for a real problem.
Closed beta · Invite only
Short daily quests put a real system in front of you. Pick the fix, order the steps, wire the architecture, then see the trade-off behind every answer.
The Architect
We'll learn system design one decision at a time.
Architecture Canvas
Try a challenge
These are the five kinds of exercise you'll do each day. Pick one, answer it and check. Nothing is saved to an account.
0 of 5 tried
Pick the best fix for a real problem.
4 islands, 12 skills, then an incident and a boss.
How it works
Each day brings a mission of a few exercises about real systems, like a slow product page or a busy database.

Every answer gets an explanation of the trade-off, and the common mix-up behind a wrong pick.

Bigger challenges ask you to connect the parts of a system yourself, then check your design against the goals.

World map
Each island holds three skills and ends with a checkpoint. Islands open in order as you go. This is an example map after the first island.
Smart Review
Smart Review picks a few short exercises on the skills you found tricky or haven't used for a while. It's optional, and it says why it picked each one.
Request an invite
Progression
Your journey shows the island you're on, the step you're learning now and the next checkpoint. Each skill shows where you stand, based on your answers.
Request an invite
Failure Lab
In the Failure Lab, a system breaks under load ("Traffic spike: database under pressure"). Read the signals, pick an intervention and see what changes. The numbers are simulated, so failing here is safe.
Request an invite
Curriculum
4 islands and 12 skills, then a Failure Lab incident and a Final Boss challenge.
Follow a request from the browser to the server and back. Learn to tell a slow system from a full one, and choose when an API call should wait for its result.
Spread traffic across servers, keep servers stateless so you can add more as load grows, and keep hot data close with a cache.
Make reads fast with indexes, copy data to more nodes for scale and failover, and decide what a reader may see after a write.
Move slow work off the request path, make retries safe with idempotency, and protect shared capacity with fair rate limits.
We're in a closed beta, so there are no reviews here yet. When beta testers share theirs, real ones will appear here.
The beta is invite only. Ask for an invite, or sign in if you already have one.
0 of 5 tried
Pick the best fix for a real problem.