How the engine selects items
- Starts near the student's enrolled grade, or from prior evidence when it exists
- Updates an ability estimate and its uncertainty after every answer
- After a missed item, probes the same skill at equal or lower difficulty to collect more evidence on the possible gap
- Ensures each in-range domain receives coverage before maximizing precision
- Randomizes among near-optimal items to limit item over-exposure
- Stops when the precision target and coverage are met, or at a maximum length
Honest about its stage
MAAT Assess currently runs an Instructional Beta engine: a Rasch-family model on a provisional, grade-anchored scale with uncalibrated, expert-assigned difficulties. It reports a provisional instructional range — never a grade-equivalent score, percentile, or national ranking. When the uncertainty is too large, it reports "insufficient evidence" instead of a number.
The path to calibrated testing
The engine is built so calibrated item parameters can replace provisional difficulties once pilot data, reliability studies, and fairness analyses support it. Formal measurement claims will be published only with documented evidence.