This dataset is derived from the existing all-data 5-core FORGE export.
Filtering policy:
Keep the existing sampled full-data iterative item 5-core result unchanged.
Reassign train/valid/test split after 5-core.
Use fixed evaluation sizes: 10,000 validation rows and 10,000 test rows.
Put all remaining rows into train.
Full user_history is retained; no recent-10 truncation is applied.
