UMSR-v1 is a curated reasoning dataset for training and evaluating models that solve problems through explicit multi-step reasoning.
The project unifies reasoning-heavy public datasets into one consistent format with provenance, quality signals, and split-ready packaging for research and model development.
