I've been wanting to do a TAS of this game for a while, so I spent a few days adapting quickerGambatte as emulation core for my bot. Shortly after starting working on this movie, GMP sent my a youtube link to TAS by TimmyAkmed and NintendoHardGamer that blew my mind.

Decided to build on their research, I contacted Timmy through Discord. He was glad to help me with some details I was missing (I only had the video, not movie files). Although all the inputs on this movie are mine, this movie is so heavily based on their route that I have to include them as co-authors.

The key glitch of this movie is left-facing wall/door clipping. On a very precise frame of the right turning action, if you input the correct buttons, collision is disabled and you can glitch through doors and gates. If you clip through a wall and fall through the bottommost screen, you enter the 'glitch' room (-1). To the left and the right are 'glitchier' rooms that contain myriad exit triggers. These are the keys to quickly skipping every level.

The main time loss in my movie is in level 6, where somehow the kid picks up a sword. This seems unavoidable and causes a ~100 frame lost compared to the old movie. I could not find whether this is an emulation difference, or it had to do with past actions in previous levels. I can live with either of them, since emulator diffs are tolerable, and my previous levels are cumulative faster than this loss.

nymx: Claiming for judging.

nymx: Re-uploading for surety on the cycle count.

nymx: Very good work....again. :P If we strictly went by frame count on this one, we couldn't say it was the fastest version...when compared against the TAS found by TimmyAkmed and NintendoHardGamer. As eien86 stated, we don't know why the optimization differences exist. Could it be emulation differences? At least I don't see anything that eien86 has overlooked. The last time I had this problem, was when DrD2k9 did a TAS of [5985] A2600 Kaboom! "maximum score" by DrD2k9 in 2:45:51.21. A youtube video was brought to our attention and showed a faster time than his submission...only the reason was discovered. Afterwards, a new movie was provided to fix the problem. So, because this is not an error that anybody can see, we would really need those inputs and the emulator used to find out why. Heck, I've seen an issue with Super Metroid ignoring lots of lag frames on elevator transitions with snes9x, where an accurate emulator has those missing frames. One thing we can say for sure, is that this record destroys the human speed run. I really like how the two show dramatic divergences, only to come back to the same ending.

Accepting.

despoa: Processing...