The problem
Many adaptive tools promise support after a wrong answer, then repeat the same item. Students learn to guess, and teachers cannot see why the path changed.
Workflow
- 01Student answers an equivalence item using large fraction models.
- 02A wrong choice is tagged with a possible misconception, shown as tentative evidence.
- 03The next item is a different, simpler supporting question with a manipulative model.
- 04A new transfer item checks whether the idea holds without the model.
- 05A summary explains the suggested next step, and the teacher can override it with a rationale.
Key design tradeoff
Transparent rules over opaque scoring
A scripted rule set is less flexible than a model that generates items, but every branch can be explained to a teacher, and one correct answer is never reported as mastery.
What you can try
- Pick 3/8 first to see the support branch.
- Restart and pick 6/8 to see the extension branch.
- Open the teacher view and record an override.
Current limitations
- Scripted items for one skill only.
- Misconception tags are plausible hypotheses, not validated diagnoses.
- No accounts, rosters, or live AI scoring.
How real impact would be measured
- Accuracy on unsupported transfer items in later sessions, not same-session success.
- How often teachers override suggestions, and why.
- Time to a correct independent response compared with a non-branching set.