A Scaffold–Engage–Apply critique workflow, a 14-type error taxonomy, and a hybrid multi-agent pipeline that scores student critiques without breaking experimental control.
Question
Students are already using generative AI on financial math. Telling them not to is not a pedagogy. The research question is whether a classroom can teach critique — of calculation, of explanation, of confidence — as a first-class skill, with the scenario held experimentally still.
Constraint
The evaluator may score a critique. It may not rewrite the student’s argument into a more convenient one. Classroom scenarios stay controlled; IRB-compliant collection is a requirement, not a footnote.
Method
CLEAR-AI implements a Scaffold–Engage–Apply critique workflow with teacher calibration dashboards. A hybrid multi-agent pipeline — math verification, an explanation rubric, and confidence calibration — scores student critiques against a 14-type error taxonomy while keeping the underlying items experimentally controlled.
Artifact
An NSF-aligned research platform spanning critique scoring, teacher analytics, and IRB-compliant data collection.
Designed and evaluated as part of doctoral research at Kean University.