**Most important category: Surface.** It has 4 values (Hard, Clay, Grass, Carpet) and is the dimension the dataset's decision use cases are built around, such as scouting and match prediction. It splits `raw.raw_kaggle` into large, well-populated groups: Hard has 117,571 rows, Clay 107,650, Grass 9,479 and Carpet 2,477. Another 28 rows have no surface recorded. **Most important metric: Aces per player-match (`Aces`).** It is the clearest single measure of serve dominance, and it moves a lot with surface. I chose it by judgment, not by a statistical test. The table and chart rendered above show the gold `raw_kaggle_by_Surface` aggregates. | Surface | Average aces per match | Median | |---|---|---| | Carpet | 7.94 | 7 | | Grass | 7.57 | 6 | | Hard | 5.88 | 5 | | Clay | 3.38 | 3 | - **Grass and Carpet are fast surfaces.** Average aces are roughly 2.2–2.4 times the Clay figure. - **Clay is slowest.** Its mean of 3.38 is the lowest, and its maximum in a single match is 41 versus 67 on Hard. - **Hard sits in between.** It has the highest single-match ace count (67) and a mean of 5.88. - **Opponent aces (`vA`) follow the same ordering.** The averages are Carpet 8.13, Grass 7.59, Hard 5.74 and Clay 3.21, so the effect is a property of the surface and not of individual servers. - **Double faults vary far less.** The averages are 2.81 on Clay, 2.97 on Hard, 3.11 on Carpet and 3.17 on Grass. The surface effect is therefore mostly about aces, not serve errors. **Caveats** - **Player mix differs by surface.** Average player rank (`Rk`) is 151 on Grass, 311 on Hard, 361 on Clay and 517 on Carpet. Part of the ace gap may come from who plays where, not only from court speed. - **Carpet is a small sample.** It has only 2,477 rows, so read its average with caution. - **Hard and Clay dominate the data.** Together they make up about 95% of the rows.