All the .snapshot/ entries makes me think this is on a B-tree file system (BTRFS); and I know from personal experience that if BTRFS snapshots are not properly handled (e.g. automatically cleaned up), they will very quickly use significant storage space. Additionally, on BTRFS, disc usage should be measured with the btrfs filesystem du instead of the plain du as the latter will show potentially wildly inaccurate usages.
Otherwise, the two biggest culprits that contribute to ballooning julias are probably always artifacts and compiled. Comparing the sizes of these directories (2358M and 4403M, respectively) with that of overall .julia (7512M), you get that artifacts accounts for about 31% of the overall .julia directory, while compiled accounts for 59%.
Now, that being said, considering that the whole math2076 directory is 213028M, Julia and all of its components make up less than 4% of the storage space while the single .snapshot/hourly.2025-08-24_1605 takes up almost 10% on its own. So it really looks like the problem has nothing to do with Julia, but rather whatever is creating and managing these hourly, daily, and weekly snapshots. That’s my thoughts anyway.