Note Column- and table-level evaluation: 27 configs (schema, join, union, grounding)

Note Row-level evaluation: row_prediction (50 OpenML tables) + record_linkage (16 entity-matching sources)

Note Compositional data-lake table enrichment: 47,772-table lake from 1,379 TabFact + WTQ parents

Note 470 MB subsample of trl-ctbench for the Dataset Large URL field