I made this test run a while ago:
https://www.youtube.com/watch?v=gzRCiFSliSA
It's not optimized as it's not using luck manipulation at all. The map is randomly generated. Some portions have a star, some don't. In order to reach the goal in fastest time, one would have to manipulate luck to get the fastest portion at all time (or at least, always one with a star). Through tests, I found that surprisingly, the RNG seems to be modified by the number of dashes and nothing else (changing starting date or number of fairies did not change the map). More tests are needed, but this would mean we'd have to slow down to get the best portion, which is sadly a bit less interesting to watch than constant dashes.
Regarding category: since this is an infinite game, a reachable goal has to be chosen for the run. I decided to go for "100 stars" at first, as going for something like "1 million points" would mean optimizing both stars and fairies, which is more complex, as taking fairies can slow you down and at some point getting both is no longer possible (see my run: I just give up on fairies at some point). When finishing the TAS, I'll probably go for something more interesting, like "200 stars". It's also possible that at some point, jumps get so long that it's no longer possible to land them, in which case "maximum score" could be an option (but, once again, much more complex to optimize).
You'll see that the score gets (comically) large, which means I may need to find a way to remove it to keep TASing (although it seems at some point the text gets so large that it's no longer on the screen at all, which solves the problem), and possibly to provide an encode without it.
Right now I've put it aside as I don't know much about luck manipulation or how to decompile flash, and even considered submitting it as-is. If someone wants to help with these, that could help me!
Edit: .ltm file for the run:
https://tasvideos.org/UserFiles/Info/638996822738108245