Answers are grounded in this paper and cite their sources.
Demonstrates adaptive sequencing improves item selection in mathematics for students, suggesting enhanced learning outcomes.
Current adaptive systems pick the next item by difficulty and knowledge-point mastery, and treat every wrong answer the same way. This breaks down in our item bank: a Grade-4 student who computes 3.25 + 1.43 = 4.68 correctly still fails ? + 1.43 = 4.68 on the same knowledge point at the same difficulty. The two items differ in cognitive direction, not difficulty. We attach a three-level state machine—STANDARD → REVERSE → ADVANCED, drawn from Variation Theory (Marton and Booth, 1997; Marton, 2015)—on top of BKT (Corbett and Anderson, 1995), gating each transition by mastery thresholds and modulating the slip parameter P(S) with a 13-code error taxonomy (careless errors get P(S) × 2.0–3.0, conceptual errors get ×0.2–0.3). On 938 simulated error logs from 12 virtual students (six learning profiles), the machine produced 23 S→R transitions, one R→A, and one ZPDprotective downgrade; a ±30% perturbation of all four coefficient tiers moved mastery by only 0.014 points weighted-average, with the conceptual-error tier near-invariant (spread 0.010). A pilot with one Grade-1–2 student (17 items, 53% accuracy) produced a diagnostic profile the parent confirmed as “about right.” The deployed system runs at 1 bit per item (correct/incorrect) under 50 ms latency; pen-input hardware would raise the ceiling to 3 bits, but we are not there yet.
No takes yet. Share an insight, caveat, or question.
yehao chen (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: