Below, columns are grouped by what they do in analysis. The profiler's own tags are sometimes wrong, because many numeric columns are stored as text. **Identifiers and keys** - The dataset has no surrogate ID column. Rows are identified by composite keys. - `players_man_.name` is the only fully unique column (uniqueness 1.0), so it is the key of that table. - `players_tournament_man_` is keyed by `name` + `year` + `tournament`, which the profiler flagged as a primary key. - The three match tables (`raw_kaggle`, `return_kaggle`, `serve_kaggle`) are one row per player per match. The key is `Name` + `Date` + `Tournament` (spelled `Tournment` in `serve_kaggle`) + `Rd`. This key comes from the ontology's reading of the data and was not checked with a uniqueness query. - `Name` and `Tournament` in `return_kaggle` are tagged as foreign keys, but the dataset card lists no verified joins. Treat that tag as a weak hint. - `raw_kaggle.against` has uniqueness 1.0, but it is a free-text opponent string, not a real ID. - `_ingestion_timestamp`, `_batch_id`, `_source_file` and `_source_system` are load metadata with a single distinct value each. They are not analytical fields. **Categorical dimensions** - `Surface` has about 4 values (hard, clay, grass, carpet). It appears in all three match tables. - `Rd` (round) has about 14 values and also appears in all three match tables. - `Tournament` / `Tournment` has about 3,963 distinct values, so it is high-cardinality. - `Score` has about 10,386 distinct values and is closer to free text. - `against` in `return_kaggle` has about 1 distinct value, so it is effectively constant there. In `serve_kaggle` it is typed as descriptive text. - `players_tournament_man_.tournament` (about 362 values) and `rounds` (about 131 values) are classifiers. - `Bpsvd` (serve) and `BPCnv` (return) are break-point saved and converted counts. They look like "x/y" text and are tagged as classifiers, but they are really metrics stored as text. - `Name` (about 491 distinct in `raw_kaggle`) is the player label. **Numeric metrics** - Stored as real numbers in `raw_kaggle`: `Rk`, `vRk` (own and opponent rank), `TP`, `SP`, `1SP`, `2SP`, `Aces`, `DFs` and `vA`. - `Rk` and `vRk` are also numeric in `return_kaggle` and `serve_kaggle`. - `players_man_.number_of_matches` is a BIGINT count. - Stored as text (VARCHAR) but numeric in meaning: - Return: `DR`, `TPW`, `RPW`, `vA%`, `v1st%` and `v2nd%`. - Serve: `Dr`, `1st%`, `1stIn`, `2nd%`, `A%` and `Df%`. - The profiler tags most of these as "classifier", which is misleading. They need stripping and casting (for example removing `%`) before analysis. - Missing data: the `raw_kaggle` stat columns (`TP`, `SP`, `1SP`, `2SP`, `Aces`, `DFs`, `vA`) are null in about 86,799 of 237,205 rows. `Rk` is null in 4,388 rows and `vRk` in 10,394. **Dates and times** - `Date` and `Time` appear in all three match tables, but both are stored as text. The dates look like `1-Apr-2013` and include a non-standard hyphen character (`9‑Sep‑2024`), so parsing will need cleaning. - The `Time` column is sparse and poorly formed, and its range is reported as "..8:30". - `players_tournament_man_.year` is a real BIGINT year (2002–2024), so it is the cleanest time field. - `players_man_.playing_years` is text holding a list of years per player.