**The most surprising finding is that the dataset looks like top-flight tennis but mostly isn't, and that ace rates differ sharply by surface. The two facts are linked.** 1. **The typical player is far below the elite.** The median player rank (`Rk`) is 237 on hard courts, 288 on clay, 109 on grass and 440 on carpet. Mean ranks run from about 151 on grass to about 517 on carpet. With ~3,963 distinct tournament labels and ranks up to 2,225, this data is dominated by Challenger-level and lower-ranked matches. It is not the ATP headline tour. Any "tennis norm" drawn from it, such as an average ace count, describes that lower-ranked field. 2. **Surface changes serving outcomes more than you'd expect.** Average aces per player-match are 3.38 on clay, 5.88 on hard, 7.57 on grass and 7.94 on carpet. Service points per match differ much less: about 73 on clay, 74 on hard, 84 on grass and 73 on carpet. - Dividing the averages gives roughly 4.6% of service points as aces on clay and 7.9% on hard. - The same division gives roughly 9% on grass and 10.9% on carpet. - So carpet has the highest ace rate even though its sample is tiny (2,477 rows) and its players are the lowest-ranked of any surface (mean rank ~517). - Double faults barely move, at about 2.8–3.2 per match on every surface. The surface effect is on aces, not on errors. 3. **A large share of the stats is missing.** `TP`, `Aces`, `DFs`, `SP`, `1SP`, `2SP` and `vA` are all null in exactly 86,793 of 237,205 rows, about 37%. These are probably matches with no recorded stats, such as walkovers or low-level events. Surface averages are therefore based only on the matches that have stats. Treat the surface comparison with caution. The ace-rate ratios are approximate, since I divided averages rather than computing per-match rates. Surface also overlaps with player level and tournament tier, which I did not control for. The result is still useful for scouting: a player's surface profile matters a lot, and a model trained on this data must account for the skew toward lower-ranked players. The figures come from the `raw_kaggle_by_Surface` gold table and the `raw.raw_kaggle` column profile.