A Multimodal Dataset for AI-Enhanced Formative Assessment of Additive-to-Multiplicative Reasoning: AMR Pilot

Team: Dr. Heidi Cian, Cheryl Tobey, Dr. Ibrahim Dahlstrom-Hakki, Rebecca Neville

Overview: The AMR Pilot project produces a public dataset focused on students’ transition from additive to multiplicative reasoning in grades 2–5. Partnering with TERC and 40 teachers, the project will collect and annotate 2,400 student reasoning episodes from formative fluency interviews.

Subject Focus: Mathematics, Grades 2-5

Targeted Universalism Focus: The project centers neurodivergent learners who express mathematical thinking through varied multimodal speech and drawing.

Public Goods & Deliverables:

  • Multimodal Dataset: 2,400 task-level records of de-identified transcripts, reconstructed student work, and teacher observation metadata under CC BY 4.0 (restricted access for raw iPad traces).
  • Documentation & Codebook: A DDI-compliant codebook, data schema with stable linking identifiers, and an Evidence-Centered Design annotation manual.
  • Proof-of-Concept Benchmark: A single-item benchmark task testing model evaluation of student reasoning quality.
  • Public Repository: An OSF-hosted package containing collection protocols, processing workflows, and starter notebooks.

Return to see all grantees.