**The most surprising finding is that surface changes how often a player aces but barely changes how often they double-fault.** Average aces per player-match by surface (from `raw.raw_kaggle`, step 8 and the gold table `raw_kaggle_by_Surface`): | Surface | Avg aces | Avg double faults | |---|---|---| | Carpet | 7.94 | 3.11 | | Grass | 7.57 | 3.17 | | Hard | 5.88 | 2.97 | | Clay | 3.38 | 2.81 | - **Aces vary a lot.** Carpet averages about 2.3 times the aces of clay (7.94 vs 3.38). - **Double faults barely move.** They stay within roughly 2.8 to 3.2 per match on every surface. A server takes more risk on fast surfaces, but the rate of errors stays nearly flat. **Carpet and grass stats come from different kinds of players.** - Carpet has the highest aces (7.94) but also the weakest field. Its average ranking is 517, against 151 on grass and 311 on hard. - Grass has by far the strongest field: its mean rank is 151 and its median is 109. - Each of carpet and grass has under 10,000 player-matches (2,477 and 9,479), versus 117,571 on hard and 107,650 on clay. Carpet's high ace figure therefore rests on a small, lower-ranked sample. - Lower-ranked players hit more aces on carpet than the elite players on grass do. Surface speed, not player quality, looks like the main driver of aces, although grass and carpet are also the surfaces with the least data. **Caveats** - About 86,800 of the 237,205 rows have no serve statistics (null `TP`, `Aces`, `DFs`). The ace and double-fault averages cover only matches that had stats recorded, while the `player_matches` counts include all rows. The aces averages are therefore probably biased toward matches and tournaments that were tracked. - These are raw averages across many tournaments and ranks. They do not control for player quality or tournament level. - 28 rows have a blank surface and are excluded from this comparison.