If something's not covered here, tell us and we'll add it.
A practice tool for system design interviews. Instead of a static checklist, you go through a full interview loop: ask clarifying questions before you're given any constraints, build an architecture on a real canvas, survive a live incident mid-build, then defend your specific tradeoffs to an AI "Staff Engineer" before getting a score.
Story Mode is a ~3-minute narrative quiz — no drag-and-drop, you tap through a crisis and the architecture builds itself as you make correct calls. It's the fastest way to see what the product does. The main simulator (from the Dashboard) is the full loop: you place and connect every component yourself, and the AI grades the actual design you built, not a multiple-choice trail through one.
No — you can explore a real curated scenario at any time from the homepage, with the actual canvas, sidebar, and scenario brief, no account required. The canvas stays locked at Clarification until you ask your first question, which is the one moment that needs a real sign-in (it triggers an actual AI call). Story Mode is the other no-signup option — a faster, narrated ~3-minute version instead of the full canvas.
Scores come from an AI model reviewing your specific architecture graph against the scenario's stated and hidden constraints, then cross-examining your tradeoffs before scoring anything. It's a genuinely useful practice signal, not a certified evaluation — treat it the way you'd treat feedback from a knowledgeable friend doing a mock interview, not an official verdict on your engineering ability.
How strictly you're graded, not just how hard the scenario is. Mid-Level rewards a working design that satisfies the stated constraints. Senior additionally expects explicit reasoning about tradeoffs and failure domains. Principal holds the design to organizational and cost-aware judgment — multi-region thinking, operating cost tradeoffs, blast-radius containment — not just whether the constraints are technically satisfied.
A toggle (the graduation-cap icon in the canvas header) for anyone newer to system design. Turn it on and the AI defines jargon inline the first time it uses it — the questions and critique stay exactly as rigorous, just in plainer language. Every component in the sidebar also has a hover/tap explainer regardless of this setting.
Story Mode needs no account at all — anyone can try it, anonymously, at any time. Signed in on Free, you get 2 sessions a month of the full interview loop (clarify, architect, defend) across Mid & Senior scenarios — including typing your own custom use case, not just the curated tracks. Pro doesn't gate custom prompts behind a paywall; it adds Principal-tier scenarios, Chaos Roulette (an AI-picked random scenario instead of one you describe yourself), and raises your monthly sessions from 2 to 30. See the pricing section for current details.
Yes — invite a friend from your Profile page. Once they sign up through your link and finish their first graded session, you both get +1 session added to that month's quota. You can stack this by inviting more than one friend (up to +3 a month); a referred friend's own bonus is a one-time +1.
Yes — cancel anytime from your Profile page; you keep Pro access through the period you've already paid for. A Season Pass has nothing to cancel (it's a single charge with no recurring billing), but the same page tells you when it expires. Full details, including what counts as a billing error we'll refund, are in the Refund & Cancellation Policy — or email support@chaosbench.com directly for a billing issue.
You own what you submit. We process it (including sending it to our AI providers) solely to run the evaluation, and may use anonymized, aggregate data to improve the product — never your individual designs shared as-is. Full details in the Terms of Service.
Yes — start a new session on the same scenario type anytime, subject to your plan's monthly session count. Each attempt is graded independently.
We'd genuinely like to hear it — send it here. Concrete reports (what you did, what you expected, what happened instead) are the most useful kind.
Ready to see it for yourself?
Design, break, and defend a real system on a free scenario — no account needed.