**The most counterintuitive finding is that the surface effect on aces looks huge, but it is tangled up with who plays where and with heavy missing serve data.** - **Aces by surface:** Average aces per player-match are 3.38 on clay, 5.88 on hard, 7.57 on grass and 7.94 on carpet (step 9). Clay produces less than half the aces of grass or carpet, which would normally be read as a pure surface effect on serving. - **Ranking differs by surface:** The players in each surface's sample are not comparable. The mean ranking on grass is about 151 (median 109), versus about 361 on clay (median 288) and about 517 on carpet (median 440). Grass matches are dominated by top-ranked players, who tend to be big servers. Clay and carpet matches skew toward much lower-ranked players, many of them from lower-tier events. The dataset has about 3,963 distinct tournament labels. So part of the "grass serves big" story is a selection effect, and the data can't separate it from the surface effect. - **Serve stats are often missing:** Serve stats (aces, double faults, points played) are null in 86,793 of the 237,205 rows in `raw_kaggle`. Coverage is also uneven by surface (step 9): - Carpet has serve stats for only 479 of its 2,477 player-matches, roughly 19%. - Clay has 63,389 of 107,650 (about 59%). - Hard has 78,342 of 117,571 (about 67%). - Grass has 8,196 of 9,479 (about 86%). Carpet's record-high ace average therefore rests on only 479 observations. - **Double faults barely move:** They are nearly flat across surfaces, at 2.81 to 3.17 per match, even though aces vary more than twofold. Serving power appears to shift with surface, but serving error does not. **Caveat:** This is observational data, so the ace gap is not a clean causal effect of surface. A fair comparison needs to control for player rank, and for missing-stat patterns, before using surface to scout players or predict matches.