data/users.jsonl: simulated user metadata and seed profiles (one user per line). data/episodes.jsonl: one row per user-day episode. data/papers.jsonl: deduplicated paper metadata with arXiv abstract/PDF URLs. data/episode_labels.jsonl: episode-paper labels, shown flags, and simulated reading selections. data/drift_timeline.jsonl: interest-drift diagnostic timeline.… See the full description on the dataset page: https://huggingface.co/datasets/OpenRaiser/PaperFlow.