Eedi - Mining Misconceptions in Mathematics

Eedi - Mining Misconceptions in Mathematics is a labelled dataset released for a Kaggle competition run in 2024 by Eedi, a UK education technology company, in partnership with Vanderbilt University and The Learning Agency Lab, with backing from the Bill and Melinda Gates Foundation and others. It draws on Eedi's bank of diagnostic multiple-choice mathematics questions, where every incorrect option (a distractor) is written to capture a specific, recognisable student error. The public data reports roughly 1,869 questions and a catalogue of about 2,587 named misconceptions, and the modelling task is to match each distractor to the misconception it reveals. You access it through Kaggle: a free account is required and you must accept the competition rules, after which the files (CSV tables linking questions, answers and misconception identifiers) can be downloaded. That gating makes availability limited rather than fully open. For RAG, the value is a clean, structured mapping between wrong answers and diagnosed reasoning errors, useful as a knowledge base for retrieval over common mathematical mistakes rather than over solved problems. For an AI tutor the fit is more specific: the dataset supports diagnosis of why a wrong answer was chosen, not merely whether it was wrong, so it can drive formative feedback, targeted hints, mastery modelling and misconception tracking across a topic. The named misconception catalogue is itself a reusable teaching taxonomy. The licence is Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0): attribution is required and commercial use is not granted, so a commercial tutoring product cannot use it without separate permission from Eedi. There is no ShareAlike or NoDerivatives condition, so adaptations are permitted for non-commercial work. Treat commercial use as prohibited under the stated licence. It complements maths sources we already list. Where amps, megamath, stackmathqa and naturalproofs supply problems, solutions and proofs, and openstax, ck-12, siyavula and khan-academy supply curriculum content mapped loosely to common-core, Eedi is distinct: it is diagnostic items labelled by the exact misconception each error reveals, aimed at pedagogy and assessment rather than content delivery.