Drift Preview 5K is a public preview dataset of 5,000 high-quality reasoning samples, curated using evidence accumulation analysis and multi-factor quality scoring.
This dataset is released under CC-BY-4.0 for research and educational purposes. For the full proprietary dataset with complete annotations (per-step loss weights, DDM trajectories, premium composite scores), please contact for commercial licensing.
Dataset at a Glance… See the full description on the dataset page: https://huggingface.co/datasets/DJLougen/drift-preview-5k.