Phase I camel-CN: curating a large-scale publicly available multimodal dataset of student actions in illustrative math to study mathematical misconceptions

The National Science Foundation Division of Behavioral and Cognitive Sciences awarded Worcester Polytechnic Institute $1.499998 million on July 1, 2026, under the Social, Behavioral, and Economic Sciences program (CFDA 47.075) to develop and publicly release a large-scale multimodal dataset documenting student reasoning errors in mathematics.

WPI will curate fine-grained annotations of student problem-solving behaviors, reasoning processes, performance, and instructional context drawn from a free online mathematics platform serving over 500,000 students across grades 3–12. The dataset spans multiple years of learning interactions and will enable longitudinal analysis of how mathematics reasoning errors emerge, evolve, and resolve in authentic classroom environments. WPI will release annotated datasets annually under open licenses and conduct dissemination through webinars, infographics, and practitioner briefs.

The award period runs through June 30, 2029, with place of performance in Worcester, Massachusetts. The project includes planned sub-awards. Assistments Foundation Inc., operating the ASSISTments platform that supplies the underlying student data, leads dataset scope definition, provides secure access to interaction logs and multimodal student responses annually, recruits and manages five teacher annotators, co-designs the annotation rubric, manages annotation workflow and inter-rater reliability reporting, supports privacy audits and de-identification, and publishes each annual dataset to Hugging Face. All data handling follows ASSISTments' existing privacy practices and FERPA, COPPA Safe Harbor, and SOC 2 Type 2 certifications.

Abstract

This project advances research on mathematics teaching and learning by developing and releasing a large publicly available dataset focused on student reasoning errors in the mathematics domain alongside related instructional classroom context. This resource aims to enable new research and discoveries of how mathematics reasoning errors emerge, evolve, and are resolved over time in authentic learning environments. The project leverages data collected in a free online math platform that includes classwork, homework and assessments across a range of mathematics concepts in grades 3-12. The platform gives students instant feedback and provides data to teachers to guide instruction. The investigators develop fine-grained annotations of this rich dataset to capture student mathematics problem-solving behaviors, reasoning processes, performance, and instructional contexts for over 500,000 students, enabling longitudinal analyses of learning interactions and outcomes across grade levels. The resulting large-scale, multimodal, multi-year dataset with detailed annotations of student learning interactions and outcomes is expected to enable future discoveries of ways to support mathematics teaching and learning. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are planned for this award.