Apex SAT

The edges in the graph are prerequisites, and they're what make it a course instead of a question bank. A skill stays locked until everything under it is mastered, so you never drill something you can't do yet. Here is one real slice of the graph, the quadratics skills. Vertex form needs standard form. Finding where a parabola peaks needs both vertex form and factored form, and factored form reaches across into the factoring skills. Some links cross subjects entirely: completing the square for circle questions builds on vertex form.

A course like this lives or dies on its questions, and 111 skills need thousands of them. I set up a content pipeline that ran in parallel, one AI generator per question type, each writing SAT-style practice for its skills, mostly at the hard end. Then a separate verifier checked their work. For math it re-solved questions independently in Python and confirmed there was exactly one correct choice. For reading it checked that one answer, and only one, was defensible from the passage.

Figures were drawn in code rather than by an image model, because image models garble numbers and labels, and the verifiers rendered the figures to inspect them. Video scouts found a short clip for each skill and confirmed it could actually be embedded.

Mastering a skill once isn't the same as keeping it until test day, so mastered skills come back for review on a schedule modeled on Math Academy's, a math program known for its spaced-repetition research. Each review you pass pushes the next one further out: one day, then about two, five, eleven and twenty-three. Because skills are linked, passing a harder skill also gives partial credit to the simpler skills inside it, so you aren't re-drilled on what you just proved. Failing a skill brings it straight back and moves up the reviews of the harder skills that depend on it. When several reviews are due, the dashboard picks the ones that cover the most others.