This is a tool assisted speedrun demonstration of Banjo Tooie 100% in 3:40:53 from power on to final collectable.
For the purpose of this (and all other real-time) speedrun, 100% is defined as getting all collectables on the totals screen for a single file: notes, jiggies, pages, honeycombs, moves, standard glowbos, and jinjos. The final boss Hag1 must also be defeated, but not necessarily at the end of the run. Stop n Swap items and the Mega Glowbo are not on the totals screen, and so are not required.
This run uses a "clean cartridge" ruleset, meaning there can't be any unlockables or save data on the cartridge at the start of the run. Real time runs on speedrun.com do not use this ruleset, and warp to the final boss fairly early in the run using a glitch with the replay menu, saving them 12+ minutes over if they started without pre-unlocks. Clean cart is the default standard in the TAS community, and one I personally believe is a better representation of "beating a game as fast as possible" (if you have to do hours of setup before you start your timer, then that isn't as fast as possible).
The emulator used for this run, BizHawk, has its share of problems running this game. The most obvious is the occasional sound glitchiness, for this run most notably in Pine Grove and in the Western Zone in WitchyWorld. But also all loading zones take longer than they would on console, costing a lot of time over an N64 run. On the flip side, the game is somewhat less laggy on emulator, so its hard to say if it is overall faster or slower. Because of the inaccurate lag emulation, optimizing for lag reduction wasn't a priority for this run.
This was originally going to be classified as LOTAD (Low Optimized Tool Assisted Demonstration), but 300k rerecords later that classification felt inaccurate. Just as much care was put into saving time as any of the other runs people call TASes for this series, even if improvements are likely possible.
Just like all other Banjo tool assisted demos since 2020, this run utilizes position microrounding codes to make the various ground clips easier. These clips are possible without them, but their otherworldly precision (to a millionth of banjo's normal movement distance in a frame), means attempting to do all 100+ of them manually would have added years to the production of this for no visible difference in the run. Between that and splicing some segments together to fix some emulation audio bugs, this isn't a "true" TAS in the sense that it is a single input file that could be submitted to TASVideos. But from a video perspective it should be completely indistinguishable from what a "true" TAS would look like.
For an in-depth on bit clips (the ground clips), see this great video: • Banjo-Kazooie - Bitcli...
The route changed a lot over the production of this, as new tricks and ideas got found. The biggest change that should have been made in retrospect would be to remove the solo section at 1:15:15 and instead do most of that in the next visit at 3:14:10, which could save 10-20 seconds. At the time of writing this, all other known minor route improvements, such as moving the cliff top jinjo to a different point, would save less than 5 seconds combined.
Some really weird stuff happens 3h21m into the run onwards. This is all to get the textbox at 3h26m50s, which is an important trigger to set up for the glitch at 3h28m20s.
This run was done entirely by myself, RingRush. However, I stand on the shoulders of giants...and normal sized people, and probably some short people. FalconXFalcon/pjii's 100% RTAs and Hyperresonance/KaptainKohl's 70 jiggy TAS were both great references to make sure I wasn't missing optimal strategies. Kohl, Hyper, and dango_v_starfish helped check over my work when routing this run, and the glitches and tech for this run that I didn't find were found by the8bitbeast, kohl, hyper, Cronikeys, Exchord, KlydeStorm and probably many more. 8bit, isotarge, ThatCowGuy, and pfedak also helped with some of the scripts that were invaluable in planning this run and discovering more tricks.
13 апр 2024
