Submission #5625: Isotarge & Mittenz's N64 Super Smash Bros. "Break the Targets Board the Platforms" in 08:05.92

Super Smash Bros. is a fighting game released in 1999 for the Nintendo 64, selling 5.5 million units worldwide. It still has a large and passionate player-base almost 20 years later, with hundreds of competitors gathering for frequent tournaments around the world.

Even though the game focuses heavily on multiplayer, there is still single player content including timed obstacle courses called “Break the Targets” and “Board the Platforms”. In this run, we complete all 24 of these courses in the shortest possible time.

The run starts from existing SRAM and a verification movie is provided. The verification movie unlocks all characters, then clears the bonus scores so that the timed portion is entirely focused on the bonus stages, making the run much faster paced and more entertaining. The inputs for the verification movie can be found here

Due to version differences the fastest version of the game for any given stage varies unpredictably. We ended up TASing all 24 stages on both the US and Japanese versions of the game to make sure we tested everything thoroughly. For submission, we chose the Japanese version which has the faster total game time by 2”16 when taking all 24 stages into account. We have made the slower but still very entertaining US run available on Mittenz' YouTube channel for your enjoyment. The video for the US run can be found here and the inputs for the US run can be found here with the US SRAM verification movie available here.

It is 2 frames faster to complete Board the Platforms before Break the Targets.

Even though we spent some time optimizing this run for real time, we would prefer the time listed on the published material to be the total game time for all 24 stages: 6'00"20 as this is much more meaningful to the Smash community than the real time length of this run.

A game time of 0”00 is possible for all 24 stages if a pause glitch is used, however this completely kills the entertainment value of the run and greatly increases the length of the run in real time. Pausing and unpausing as fast as possible is very distracting, and takes much longer to complete the stages in real time. For these reasons we have chosen not to use this glitch and would advise that any future improvements to this run follow this precedent.

There is a long history of TASes for these stages over the past decade, before we started this project the total time for all Break the Targets stages (Japanese and US best times combined) was 2'07"56, after our improvements the combined best times now sit at 2'06"72. The total time for all Board the Platforms stages (Japanese and US best times combined) was 3'57"15 before we started this project, after our improvements the combined best times now sit at 3'51"31. Throughout this project we have saved a total of 6”67 when combining the best known times of the US and Japanese versions. We finished with a total of 204,122 rerecords for all 24 stages on the Japanese version over roughly 8 months.

The competitive nature of the Smash 64 community had spilled over into the TAS scene for this game, it was extremely rare to see inputs posted publicly for these stages when a new record was set. The only resources we had to compare against while working on this run were YouTube videos of the records, often at 30FPS. As far as we are aware, this is the first time that optimal inputs have been publicly available for all 24 stages on both versions. We are excited to see more TASers attempt to improve these stages now that all of the resources are publicly available, and we hope that the resources remain open as improvements are made.

This stage is cycle based until after the moving lava platforms. We finish behind the best known time for the US version, 15”19, due to shorter distance covered by Mario’s up-special.

The second half of this stage is cycle based. A time of 16”25 is possible on the US version due to Yoshi’s faster aerial movement speed.

Another cycle based stage, the Japanese version loses some time at the start due to differences in Smash DI, but catches up to the US version in the end because they both make the same platform cycle.

The start of this stage is cycle based, we saved some time before the cycle but that was nullified by the longer wait for the platform that extends from the wall. Smash DI is used to gain extra distance from the bumper damage boost. The best known time on the US version is currently 26”83 due to greater distance travelled by Smash DI.

It was originally thought that the US version was faster than the Japanese version for this stage due to Link having higher gravity and better smash DI. However, in the Japanese version it is not necessary to tech or edge cancel damage boost off of Link’s bombs due to them having less knockback. The best known time on the US version is currently 17”70. The European version beats them both with the best known time currently at 16”55 due to increased horizontal aerial velocity.

This stage was thought to be capped by a cycle at 24”99, but one night we decided to reroute the stage completely to try and remove the cycle from the equation, this was successful and over the next few weeks, we lowered the time by over 2.5 seconds. The best known time on this stage for the US version is currently 22”10.

This stage is cycle based up to the 6th platform. The Japanese version loses time at the bumper damage boost compared to the best known time on the US version, 23”31, due to 40% less distance travelled during Smash DI frames.

This stage is quite unique, with much of the time spent zipping around in Pika’s up-special. The US version can match the Japanese time to the frame, none of the version differences affect this stage.

Smash DI is used to ensure the correct angle from the bumper damage boosts. Originally we thought that the end of the stage was cycle based, we’d start the initial dash from very far right on the falling platform and wait for the platform to land so that the running animation wasn’t interrupted by running off of the edge of the platform. Thankfully we improved the start and mid sections to the point that it actually worked out faster to fall off of the falling platform before starting the final initial dash from the ground instead of the platform. Teleports are very useful in this stage, they can be done by switching running direction for a frame, and 3 frames later performing an action such as jumping. This causes a massive horizontal boost for a few frames. The best known time for this stage on the US version of the game is currently 20”39 due to the Smash DI advantage.

We rerouted part of the stage to eliminate backtracking, saving over a second on the previous best known time of 20”83. This is one of the more simple stages on paper (lots of running in straight lines) but we ended up spending over 14k rerecords on this stage. The up-special that enables this route to work appears to be impossible on the US version due to differences in the jump physics, giving a massive advantage to the Japanese version for this stage.

We improved the previous best known time of 20”69 significantly on this stage by cancelling the final PK Thunder landing early. The best known time on the US version of the game is currently 19”65 due to the greater distance covered by Smash DI enabling us to use an extra two damage boosts at the end.

This stage is cycle based due to the swinging platform, our starting route is several frames faster than the reference video for 19”15 but we’re forced to waste some time before jumping up to the pendulum platform. The Japanese version is faster for this stage, the best known time for this stage on the US version is 19”23.

This stage is faster on the US version, with a best known time of 7”95 due to the greater distance travelled by Mario’s up-special. Unfortunately this stage is also capped by a platform cycle.

The best known time for this stage on the US version is currently 14”10, unfortunately the Japanese version loses some time due to lower horizontal velocity when moving through the air and other minor physics changes.

Most of the time save for this stage comes from carefully timing the final down air so that the largest hitbox comes out on the first frame that it would connect with the target. The best known time on this stage for the US version is currently 8”65.

One of the most difficult stages to optimize, many hours were spent getting the boomerang shot at the end to connect with all of the needed targets. Several alternative routes were tested with none saving time. The best known time for this stage on the US version is currently 10”39. The European version has a large advantage on this stage due to a boost in aerial speed, with a best known time of 9”85.

This is another stage that was thought to be capped by a cycle at 8”75, however with careful optimization of the start and a small reroute for the ending we were barely able to make an earlier cycle and save 0”35. Because of an extra freeze frame on the US version every time direct contact with a target is made, the fast cycle is narrowly out of reach on the US version. Interestingly if an aerial laser was a single frame faster a time of 8”3x could be achieved with a slightly different route.

This stage is capped by a cycling platform in the center of the stage, we land on the platform sooner than previous runs, and use this time to charge a shot slightly. We were hoping that the added size of the charged shot would allow us to shoot to the left from lower and thus land at the bottom sooner but sadly this didn’t end up saving any time. The best known time for this stage on the US version is currently 8”90 due to extra freeze frames after the platform cycle.

This was the first stage that Isotarge TASed back in December of 2016 after competing with The8bitbeast in real time attempts. Not much to say about it other than it was a fairly easy introduction to this game, several important mechanics were learned without the stage being prohibitively difficult to optimize. The previous best known time for this stage was 8”51 which was improved by a frame thanks to very careful optimization of the start. The input for the down-special occurs 46 frames before the stage is completed, which means it was 1 frame away from being an option for the last stage completed in this run.

We improved the previous best known time for this stage, 10”25, by 1 frame thanks to extreme optimization of the up-specials and ending. The best known time for this stage on the US version is currently 11”10 which uses a very different route.

We improved on the previous best known time of 15”01 by attacking the second target with an up-air instead of PK Thunder, and using the extra height from this jump to reduce the time needed for the PK Thunder to connect with the third target. Double jump cancels are important for this stage and save significant time when running along the ground. The best known time for this stage on the US version is currently 15”55 due to some significant version differences.

The most difficult part of this stage is hitting the up-airs without wasting time with extra height or distance before landing. The wall by the final target is not solid from the left and this allows Falcon to land back on the stage after hitting that target using an up-special, sometimes zipping horizontally by up to 800 units. We tried an alternate route which completes the stage backwards using this trick, unfortunately our test run was almost a second slower than the standard route.

This was a difficult stage to optimize, specifically the jump spacing to land on the small ledge while both hitting the target and cancelling the landing optimally. The best known time for this stage on the US version is currently 12”99 due to extra freeze frames when hitting the targets.

This stage was quite difficult to optimize, we were consistently 2 frames behind the reference video for the previous best known time of 13”40 at the start and couldn’t figure out why. Eventually we figured out that a double jump is required to land sooner on the edge and managed to save the remaining frames needed to tie 13”40. The stage was then further improved by a somewhat counter-intuitive technique: Not holding left/right on the joystick after jumping in the air. Attacking while jumping in the air has horizontal velocity of 27.5 and this velocity is preserved through jumps even if the next jump doesn’t attack. When the next jump doesn’t attack, the velocity counts down in increments of 0.5 from 27.5, however holding a direction during this countdown instantly snaps it to the velocity cap of an aerial jump without an attack, 21.9, which lowers the horizontal distance covered by the jump quite significantly. So it’s fastest to let go of the control stick completely until the velocity counts down from 27.5 to 22 before holding the needed direction to cap it at 21.9. The reason we complete this stage last is that the final input (up-special) occurs 47 frames before the stage is completed, which allows the input file to end early, saving some time when using TASvideos standard timing (power on to last input).

Special thanks to my co-author Mittenz. Thanks to The8bitbeast, Tiffanyjane, and Nicolas Frechette for pointing out several timesaves. And of course, a huge thanks to the previous TASers of these stages for over a decade's worth of reference videos, routes and inspiration: JPleal10, Haru, BlazeSSB, antdgar and any others we may have missed.

Started on BizHawk 1.11.9 (syncs on 2.1.1) played using TAStudio and ScriptHawk, pure interpreter, GlideN64 at native resolution of 320x240.

Noxxa: Judging.

Noxxa: This movie was received very well, and it is easy to see why: the Board the Platforms and Break the Target levels are very well suited to showing off all the characters' different abilities. For its eight-minute runtime it has a lot to show off, and it does so well. The movie is well optimized and the authors clearly have done extensive research (including fully TASing both regions just to be sure which one is faster) and put a lot of work into it. Accepting to Moons as a new category.

thecoreyburton: Processing.