Does adaptivity actually help?
Mentora has no real classroom deployment yet, so these numbers come from an in-browser simulation, not production telemetry — labeled as such throughout this page. What it tests is real: the exact bkt.js engine and recommendNext() logic that runs a live lesson, applied to synthetic students with a known hidden "true mastery" per concept, so we can score correctness against ground truth instead of just self-consistency.
Two strategies race to detect mastery on the same 5-concept dependency graph: random-order quizzing (fixed round-robin through concepts) vs. adaptive quizzing (Mentora's real recommendNext() — lowest-mastery concept whose prerequisites are already met). Both use the identical BKT parameters and stopping rule.
Simulation controls
Larger sample sizes take a moment longer but reduce noise in the reported averages.