Every claim on this page describes what the engine does today, not a roadmap. Where something is a genuine limitation, we say so.
A short adaptive screening test builds a starting picture across the topic — not a fixed set of questions everyone gets the same.
Every answer updates a per-skill mastery probability using Bayesian Knowledge Tracing, a published statistical model — not a black-box score.
The engine looks for a genuine pattern across several attempts and question types, not a single wrong answer, before calling something a gap.
Practice targets the specific failing skill and its prerequisite — not just "more questions on this topic."
The repaired skill is re-tested before the plan moves on. Progress is checked, not assumed.
The mastery probability behind every skill uses Bayesian Knowledge Tracing (Corbett & Anderson, 1994) — decades-old, peer-reviewed learning-science research, applied the same way in classrooms and research studies worldwide. It isn't something we invented, and it isn't a mystery: it updates on four things — whether the answer was right, how consistent the pattern has been, whether the learner has shown it across more than one question style, and whether the skill has held up over more than one sitting.
The underlying model is public, peer-reviewed research, not a secret formula. How it decides can be explained in plain English — as it is on this page.
The model currently runs on the commonly-cited generic defaults from the research literature, not parameters fitted to Grasp Maths's own students yet. We're actively logging anonymised per-answer data to calibrate that honestly, rather than claiming a precision we haven't earned.
The engine finds gaps and recommends the next useful practice. It doesn't replace a qualified tutor's ability to explain, encourage and adapt in person — the two work best together.