Adaptive Learning Engineering — BKT, IRT, CAT, routing, validation. The course runs the engine it teaches on your answers.
Modules
The engine's model of you
Decision log — every claim, with its why
How this model of you works (and where it cheats)
Per atom, the engine runs the exact BKT update you meet in Module 2. Lesson checks are practice: you see feedback, so learning credit (T = 0.25) applies after each answer. Test-outs are evidence-only (T = 0): proof, not assumed learning — Module 2, lesson BKT-005 is about precisely this switch.
Guess and slip come from format and difficulty, the Grade 3 table: easy MCQ 0.30/0.08, medium 0.25/0.10, hard 0.18/0.13; numeric 0.03/0.12; select-all 0.08/0.15. An atom is CERTIFIED at evidence P(L) ≥ 0.90 with at least two items and one non-easy correct — your own 90% bar.
The ability track is a streaming 1PL (Elo): θ ← θ + K(actual − expected), expected = logistic(θ − b), with b taken from difficulty tags (easy −1, medium 0, hard +1, +0.4 for typed/multi). The error bar is 1/√(Σ information). This is deliberately the cheat IRT-007 warns about — tag-based b, uncalibrated — shown here because an honest uncertainty label beats a falsely precise one. Everything persists in this browser only (localStorage); reset below.