HiCogMath: A Cognition-Aligned Dataset for Higher-Order Competency-based Mathematics Assessment

The dataset was created from mathematics curriculum-aligned assessment items in Uganda's lower secondary. Using established cognitive frameworks, the items were systematically screened, aligned with specific constructs, and adapted contextually to reflect real-world scenarios. Each row is structured using a multi-field schema comprising items, conceptual and linguistic features computed using NLP tools and techniques. Annotation was conducted by five expert secondary school mathematics teachers via an independent paired-review design, with the last person acting as an adjudicator. Reliability was quantified using Cohen’s Kappa for pairwise annotations and Krippendorff's alpha for the overall annotation.