Build an E-Commerce Store From Zero
Ten stages with feedback only, then the final capstone: a marketplace, chosen because it is not the store with more tables. At every stage you write your answer first and reveal what a strong answer contains second. Atlas asks questions and unlocks domains; it does not hand over an architecture.
The exercise exists so that Atlas never shows the solution first. Every stage has an answer that looks complete and is not — a stack list where a problem frame was asked for, a diagram with a message broker where a slice was asked for, “handle errors” where a failure plan was asked for. Your answers stay in this browser.
Build an online store.
That is the whole brief, and it is deliberately the whole brief. The capstone works in ten stages, and at every stage you write your answer first — a problem frame, a V1 cut, a data model, a slice, a research plan, a failure plan — and only then reveal what a strong answer contains. The reveal is feedback, not a reference: compare it with what you wrote, notice what you skipped or over-built, and revise before moving on. Nothing later in the capstone will make sense if you read the expected answer before writing your own, because the skill being trained is producing the answer under uncertainty, not recognising it once shown.
Frame, cut to V1, discover the data, build the thinnest slice, take the cart decision seriously, reach an order without payment, research payment as questions, inject failures, and only then production and scale.
Problem Frame
Write one paragraph, in the words of the person who asked for the store, saying what it is: who uses it, what each of them does, which single workflow means it works, what is outside the system, and what is not being decided yet. No technology names.