The most counterintuitive result I can support is that **surface changes aces a lot but barely changes double faults.** Part of the surface effect may also come from which players appear on each surface. This comes from the per-surface summary of `raw_kaggle` (gold table `raw_kaggle_by_Surface`), which I queried. I did not run a further test to separate the two effects. - **Aces vary widely by surface.** Average aces per player-match are about 7.9 on carpet, 7.6 on grass, 5.9 on hard and 3.4 on clay. Grass produces more than twice the aces of clay. Carpet has only 2,477 rows, so its figure is less reliable. - **Double faults barely move.** The average is about 2.8 on clay, 3.0 on hard, 3.1 on carpet and 3.2 on grass. Servers hit far more unreturned serves on fast surfaces, yet they do not make many more errors. - **The player pool differs by surface.** The average ranking (`Rk`) is about 151 on grass, 311 on hard, 361 on clay and 517 on carpet, where a lower number means a better player. Grass matches therefore involve noticeably higher-ranked players, and only 9,479 rows are on grass versus 117,571 on hard and 107,650 on clay. Some of the ace gap may come from who plays on each surface rather than the surface itself. Controlling for rank would be needed to separate the two. - **Serve stats are missing for much of the data.** `TP`, `Aces`, `DFs`, `SP`, `1SP`, `2SP` and `vA` are null in 86,793 of 237,205 rows (about 37%). Every surface average above is based only on the matches that have serve stats. I did not find a stronger or more surprising effect than these in the evidence I pulled.