This brought the WR down from 1h 12m 51s to just 46m 55s. This will probably be a separate category from the old any%.
Apparently, the Japanese community had quietly been working on this and just dropped this massive discovery out of nowhere.
From Noki Doki in the Discord
You use cutscene underflow to replay the right bell Shine spawn cutscene played previously, but since you're on a different map (Ricco/Gelato unlock plaza vs. Shadow Mario chase plaza), the right bell Shine isn't at the same memory address anymore, and the old address (which the cutscene still has) now holds something else, in this case render data for the turbo gate behind the lighthouse.
We can line up that turbo gate at a very precise position on screen so that when trying to spawn the "Shine", the cutscene will read the position as a memory address and run code from wherever we want. The game keeps track of stick angles to check for spins, which is the easiest memory area to manipulate that we know of, so that's where we write the arbitrary code
I'm always fascinated by the extreme puzzle solving behind these things, but still not super familiar with speed running. (Mostly just watch Karl Jobst.) I know what ACE means, what is RTA?
RTA means "real time attack". In this case it just means a human is capable of doing the full run instead of only being able to do it in small pieces with tons of attempts, or TAS
And for anyone else, ACE means arbitrary code execution, where you can write values into memory where they will be executed as instructions
Thanks!