**The time axis is `raw.raw_kaggle.Date` (tournament start dates, at most about 79 distinct days a year). It has one clear dip, one partial-year drop, and a long ramp-up, and its coverage shifts around 2016.** The figures are player-match rows per year from the yearly query I ran (step 1). - **Steep ramp-up, 2001–2015.** Volume rises from 42 rows in 2001 to 13,631 in 2015. The first years look like sparse, partial coverage rather than real activity: - 2001 has only 9 active days, between August and October. - 2002–2004 have 128, 292 and 342 rows. - Volume then roughly doubles every few years, reaching 1,801 in 2007, 5,715 in 2011 and 13,631 in 2015. - **Level shift in 2016.** Rows jump from 13,631 to 16,454 and then stay at 17,600–19,900 through 2019. Active days fall from 76 in 2015 to 55 in 2016 and 2017, so the data likely moved to a different collection or tournament-date granularity. More rows per date means more matches are being logged per start date, which is not obviously a real change in play. - **Pandemic dip, 2020.** This is the clearest dip: 9,253 rows, against 19,864 in 2019 and 20,533 in 2021. That is about 53% below 2019. Active days fall to 31, compared with 63 in 2019 and 76 in 2021, so the tour was suspended and then restarted. - **Post-pandemic peak, 2021–2023.** Rows rise from 20,533 in 2021 to 24,723 in 2022 and 25,369 in 2023, the highest in the dataset. - **2024 is partial, not a real decline.** The 22,689 rows stop at 2024-10-28 (59 active days), so the year is about two months short. The overall raw range of the card ends at 9 Sep 2024, which may point to slightly different cut-offs in the other tables. Two cautions for trend analysis: - **Partial years.** 2001 and 2024 are cut off. - **Uneven coverage.** The pre-2016 growth reflects coverage as much as tennis activity, so year-over-year comparisons before 2016 are unreliable. The card's ranges also show `serve_kaggle` starting in 2006 while `raw_kaggle` and `return_kaggle` start in 2013 by text sort. That suggests the three fact tables don't cover identical periods, and I did not check this further. Two other caveats: - **Mixed date characters.** Some date strings use a non-breaking hyphen (e.g. `9‑Sep‑2024`), so date parsing needs care. - **Nulls in match stats.** About 86,793 of 237,205 rows have null match stats (TP, Aces, DFs and similar). That is 37% of the table, and the share may also vary by year. {"action":"final","content":"placeholder"} Correction: the response above should be read as the final answer; the trailing lines were a formatting error.