This run originally was meant to be a simple resync to CGB-GBA mode, but ended up turning into a (questionable) improvement.
The main improvement here is the use of "moonwalking". Simply be, the game does a walking animation while you move if you hold the moving direction. However, if you don't hold that direction (i.e. tap it each time), you don't do this animation, resulting in a "moonwalk". This seems to reduce lag slightly, and even does so on the old TAS (not using latest hawk) if applied (at least as far as the first 6 levels), so this probably counts as an improvement.
However, you can note that this movie seems to be around 9 seconds slower than the old movie (disregard BIOS time, it's more 6 seconds). This seems to indicate that more accurate emulation results in more lag. As I've noted, I've tried applying the moonwalk to the first 6 levels on the old TAS and it ended up resulting in less lag (around 4 frames). I couldn't check anywhere past that due to the lua script I've used not behaving correctly on old hawk versions. I am fairly confident overall this is an improvement, but it is still somewhat questionable.
On a fun note, this TAS had 48187 botted rerecords, give or take a few hundred for lua script trial and error.
Lua scripts used:
EDIT: Lua script used for v2 of the TAS:
Also, the truncated input files used:
Also I used a dev version of BizHawk because lol this game, probably syncs on 2.6.2 (no it doesn't lol) but if it doesn't the submission can be delayed until 2.6.3/2.7 release.
feos: Delaying until new bizhawk release.
feos: New hawk is out now. Resetting.
feos: Judging...
feos: Updating the movie with an 8 frame improvement.
feos: I have to reject this movie in its current form. We've done some work on boiling this improvement down to something verifiable, but it doesn't look like it can really be verified using this new version of the core.
The main issue is that there's actually no guarantee that the new trick makes enough levels faster if we account for lag. The differences in times are all over the place: some levels are shorter, others are longer, and time discrepancy also varies greatly. Maybe in some places the new trick causes more lag than necessary. Which means the old approach has to be tested as well, for every level, and maybe even for every turn where it's an option.
If our only way to verify this is redoing the entirety of the old movie on the same old version of the core (to prove it's better in identical conditions), then that needs to be done as a separate submission, because we can't delay this one indefinitely, and the author hasn't given an estimate on when it could be finished.
Rejecting as explained in this post.