Eighteen columns are classified as categorical (`rough_kind = categorical`). Their distinct counts come from the catalog metadata query in step 3, and they are approximate (`approx_distinct_count`) rather than exact. **Very low cardinality** - `return_kaggle.against` has only 1 distinct value, so it carries no information. - `Surface` has 4 distinct values in each of `raw_kaggle`, `return_kaggle` and `serve_kaggle`. These are the surface types (hard, clay, grass, carpet). - `Rd` (round) has 14 distinct values in each of the three fact tables. **Low to medium cardinality** - `players_tournament_man_.year` has 24 distinct values, covering 2002 to 2024. - `raw_kaggle.DFs` (double faults) has 29. - `raw_kaggle.Aces` has 54. - `raw_kaggle.vA` has 62. **Higher cardinality** - `raw_kaggle.2SP` has 134 and `raw_kaggle.1SP` has 189. - `serve_kaggle.Df%` has 231 and `serve_kaggle.Bpsvd` has 236. - `Name` has 491 in each of `raw_kaggle`, `return_kaggle` and `serve_kaggle`, meaning the same set of players appears in all three. Some of these columns are really numeric measures that were labelled categorical because they have few distinct values (`DFs`, `Aces`, `vA`, `1SP`, `2SP`). Others, such as `Df%` and `Bpsvd`, are stored as text.