I worried ever so slightly that this would break rules regarding thread bumping, but I suppose this is important enough.

With all of the new strategies that have been found recently, I would appreciate some help in formally documenting them all. At the moment, new setups for different clips, flagpole glitches, etc. have been scattered all over the place, and I am unable to keep track of all of them.

The goal is to compile these resources into one place so all runners can easily access them for future reference. Thanks.

I decided to investigate ways of saving the 4-2 framerule, and I found this method, which happens to just barely be possible without any double fast accelerations.

The TAS saves the framerule with 0 frames to spare, so the execution must be extremely precise, but I think this might be feasible with a minimum of 3 segments (the two bumps, the pipe entry, and the warp zone). Four or five segments may be more realistic though.

It's slower than clipping into the top layer of bricks by a few frames, but it is much easier to perform bumps than to perform that clip. Perhaps someone will be able to perform that floor clip at some point, but we have to build up to that, it seems.

Because I want to improve the speedrunning community, I have decided to start my own server for SMB1-related speedrunning and TASing (although you are free to talk about other games as well).

The invite can be found here: https://discord.gg/wPt8twK

My goal is to improve upon the organization of the SMB1 speedrunning community and provide a better place for people to discuss strategies and help each other get started with the game. There are many server improvements compared to the existing server. To obtain Runner and/or TASer roles, you may use the Reaction Roles bot in the role-request channel.

Maru

Since this issue has been occurring time and time again and especially with this game, I thought I would raise attention to it here.

Various users, many of which are from the SMB1 RTA community, have submitted TASes of SMB1 that do not follow important guidelines on TASVideos. Some of those guidelines include:

1. The movie you submitted did NOT beat an existing record. In the vast majority of cases, the movie is not beating the existing SMB1 warps record by HappyLee (4:57.31). Human theory TASes are also not suitable for TASVideos publication.
2. The movie you submitted involves a category that offers very few differences from the existing categories. Examples of this include the SMB1 no left+right category, which does not offer many notable differences compared to the conventional warps category.
3. The movie you submitted involves the use of cheat codes. The use of Game Genie codes is forbidden for submissions.
4. Your movie features noticeably suboptimal play. A decent amount of effort is required to be put into submissions. For example, if you are bumping into walls or (for this game specifically, not performing a flagpole glitch) for no good reason, that is a sign of suboptimal play.
5. You have submitted a TAS of an SMB1-based ROM hack that does not offer unique features compared to the original game or other existing SMB1 ROM hacks.

For reference, you can look at these links to see the long list of non-published SMB1/SMB2J submissions. Chances are that they were rejected for one of the reasons that I listed above.

http://tasvideos.org/Game/nes-super-mario-bros.html http://tasvideos.org/Game/fds-super-mario-bros-2.html http://tasvideos.org/Game/fds-super-mario-bros.html

I am not discouraging people from showing off their work, but if they would like to show off their work, they should use websites such as YouTube or the TASVideos userfiles section instead (http://tasvideos.org/userfiles/). If you are planning on submitting an SMB1/SMB2J TAS for publication to TASVideos, make sure that it follows the guidelines. In addition, I recommend that you guys share your WIPs so others can review it.

Thank you everyone. If you take what I have put here into consideration, it will make our work easier.

Maru, TASVideos Judge

Mars02: "Saving subpixels? I don't think you "save" subpixels and you don't save time by doing so, if it's possible."

If one more subpixel could be saved in the 8-4 walljump room, the SMB1 warps TAS would be one frame faster.

Regarding the clip, another frame might be possible to squeeze out of the TAS, but it would require some tight optimizations to be able to stop on the perfect subpixel one frame earlier.

I decided to look into finding setups that could possibly be done in real time for clipping into the top floor of 4-2. Here was one that I found in a few minutes of testing:

http://dehacked.2y.net/microstorage.php/info/315575917/SMB14-2TopClipSetup.fm2

It does not require any one frame right releases or switching between right and left in a single frame, but it does require a two frame right release in order to manipulate subspeed values. Note that I did not play through the entirety of 4-2, but it is fast enough to save the framerule if everything else is executed well. You can use the frame advance key to see the inputs that I used.

Obviously, this trick is still going to be very difficult. There are still a lot of moving parts involved in clipping through the bricks. I think that people should share setups that they find here so that it becomes a collaborative effort to find the best way of clipping into the floor in this way.

After an extended period of time, the RTA Rules TAS has finally been improved. The new time is 4:57.54 from power-on, which converts to 4:54.26 using RTA timing.

The improvement comes from the turnaround room in 8-4 and stopping on the floor one frame earlier while still being able to scroll the screen far enough for the wrong warp to work. The improvement is very precise and depends on subspeed values ($705) because if the subspeed is not optimized, then it is impossible to scroll the screen far enough.

Not all levels were completely optimized for speed. There is a small improvement to 1-2 and several frames can be saved in 4-2, but those improvements do not mean anything due to the framerule. I also did not bother to optimize subpixels upon performing the 8-4 walljump or accelerating towards the pipe in the turnaround room.

Here is a video: Here is the input file: http://dehacked.2y.net/microstorage.php/info/315676946/Maru_SMB1_RTARulesTASImprovement.fm2

Over the past few days, I've been looking at 1-2 in order to see how much I could squeeze out of it. I think I have reached or come close to the limit. I was able to finish 5 frames away from the next framerule. For reference, this is 3 frames faster compared to HappyLee's warps TAS, but that TAS also sacrificed a frame for entertainment purposes. Here is my perfect 1-2 TAS: http://dehacked.2y.net/microstorage.php/info/1398002361/SMB1Perfect1-2.fm2 . It may be possible to save another frame in 1-2, but unfortunately, reaching the next framerule does not look to be possible.

It is very interesting how the pipe clip and wall clip methods in 1-2 are almost equal in speed. However, the pipe clip seems to be slightly faster. It is also possible to be fast enough for the optimal framerule without using any clips. Not too many people know that.

This strategy is slightly different than strategies of turning around in the warp zone in previous TASes, and it is way more precise. Previous TASes used a jump turnaround that was similar to the turnaround strategy used in 8-4. This TAS instead relies on a complete fast acceleration to turn around quickly. Upon performing the pipe clip, Mario's movement has to be manipulated in order to not pass x-position ($3ad) 125. With this x-position, it is possible to make a subpixel-perfect stop on the second pipe (stopping on the last subpixel of the pipe) and begin turning around to the World 4 pipe.

I decided to experiment with 4-2 a little more today. Still, it seems very unlikely that the final frame in that level will be saved.

Here is a TAS that I put together of the level using the most optimal routes and movement: http://dehacked.2y.net/microstorage.php/info/1138750230/SMB1Perfect4-2TAS.fm2

Doing the wall clip requires subspeed manipulation and specific movement in order to push Mario's screen position more. If the movement is not optimal, frames will be lost performing the clip and getting to x-position 132 ($3ad).

The one frame improvement to the warp zone compared to HappyLee's 2011 warps TAS (and every other speed-oriented SMB1 warps TAS from late 2004 onward) is barely possible and was subpixels away from not being possible.

I hope that the next frame is found, but I have my doubts about it being possible.

It is worth noting that the framerule save in 4-2 is possible with fire flower by using a faster clip, but unless it is possible to somehow despawn one of the piranha plants, it probably holds no use for any%.

This post is concerning the NTSC version of the game.

The TAS limit seems to be 4:57.31 (4:54.03). While it is bold to say that a TAS is perfect, 8-4 of SMB1 has only been improved by two frames since late 2004.

That TAS uses left+right, but without left+right, the fastest the game can be beaten at the moment is 4:57.56 (4:54.28). I am less certain that this TAS is perfect, however, because when I made my version of this TAS, I was close to saving a frame in the third room of 8-4 with more optimized subpixels ($400) compared to HappyLee's version. If the TAS can be improved, there would only be one frame to save.

Here is a breakdown comparing human perfection to the RTA Rules TAS:

1-1: Human perfection has four frames to spare, while the RTA Rules TAS has 16 frames to spare. 1-2: Human perfection has six frames to spare. The RTA Rules TAS has 11 frames to spare through a humanly impossible clipping strategy, however. 4-1: For RTA speedrunners, there is actually only one frame to spare in this level. The TAS, on the other hand, has about 9 frames to spare. 4-2: The TAS can save several frames through more optimized movement, ultimately resulting in saving a framerule. 8-1: Most speedrunners finish the level with no frames to spare. It is possible to save a framerule by performing a fast acceleration and flagpole glitch, saving the framerule with one frame to spare (an extra frame is saved walking to the castle). However, this is rather unrealistic. The TAS has two frames to spare in this level. 8-2: The TAS saves another framerule with more optimized movement and better luck manipulation. My TAS had two frames to spare in 8-2, but it may be possible to save a couple more with a better bullet bill shot and more forensic searching for better luck manipulation. The timer for the bullet bill cannon needed for the bullet bill glitch is located at $480. 8-3: Both the TAS and the RTA record have to wait for the timer to hit 242 before performing the flagpole glitch. 8-4: The TAS saves several frames over both human perfection and the world record in 8-4. I don't know how much faster it is compared to the human theory TAS, but according to Bismuth's video, it was 28 frames faster than somewes' world record.

MyLittleWalrus asked me to post my opinion in this thread, so I've decided to go ahead and do it. Keep in mind that I am mostly involved with the SMB1 and SMB3 communities, not NSMBDS, but I have talked with a variety of runners from this community.

The first issue is emulator accuracy. Desmume is not a very accurate emulator. These emulators are not accurate to the point that they will console verify. Relative to an actual console, Desmume is not as accurate than other emulators such as FCEUX, BizHawk, and lsnes. Times on emulators would appear quicker than times achieved on DS consoles. I am aware this holds true for the Wii Virtual Console, so that also has to be considered.

The next issue is cheating. I think NSMBDS is lucky to not have as many cheaters as other communities such as SMB1 and SMW. With emulators, it is not hard to record movie files using rerecords, frame advance, and slowdown features. I think a simple solution to this is to encourage streaming attempts on Twitch or YouTube. Some communities strongly recommend input displays, but streaming history is important for gaining trust within the speedrunning community.

Do I think emulators should be banned? No, but the DS does not have accurate emulators compared to the NES or SNES. This is a valid reason as to why certain emulators should be disallowed, however, but given that there aren't any better alternative emulators to Desmume, I suppose it's fine to allow it as long as these discrepancies regarding timing are clearly noted. When DS emulation becomes more accurate, the best course of action would be to deprecate older, inaccurate emulators.

Maru

I'm going to bump this to fill in with some more information. At the time of this post, the current world record is held by MitchFlowerPower with 10:44.

Mitch may incorporate autoscroller clips into future attempts of this category. He's allegedly been having solid success with the autoscroller clip on 8-T. There are 3 possible places to do autoscroller clips, but the one in 8-T2 is not RTA viable. There is also a setup for an 8-Navy autoscroller clip that Kirua gave me a link for:

Each clip saves about 40 frames if done perfectly. I don't know if that was factored into the theoretical theory TAS time, but it was never specifically mentioned.

PAL is not your pal.

@ CuBored

Nobody has saved the 4-2 framerule with HappyLee's method yet.

Never did I say the run was dead. There are still several frames to save in 8-4. But there is only so much a human can do in succession. If 8-1 was the first level in the game, perhaps we would already be seeing people attempt to save the framerule, but it is not. Saving that framerule in attempts asks for too much at the moment, as an RTA speedrunner would have to perform a perfect fast acceleration with perfect inputs, manipulate subpixels (praying that you did the "perfect fast acceleration"), and perform a more inconsistent flagpole glitch.

Hello, I suppose I should introduce myself since this is my first thread on this forum. I'm Maru, an SMB1 and SMB3 TASer. I've done quite a bit of work with both games, but I decided to come here to address some of the issues with speedrunning this game.

A while back, I published an RTA Rules TAS showcasing some different strategies to save the framerules on 1-2 and 4-2. A number of people have requested that I post the input file, so here it goes: https://cdn.discordapp.com/attachments/351570852308975620/497923454369267712/Maru_RTA_Rules_TAS.fm2

This TAS aims for pure speed and highest score (without losing time of course). It's not incredibly entertaining, but a surprising amount of my movement was in sync with the music.

Using Sockfolder values, the strategy on 1-2 saves 5 frames (4-D) over human perfection (4-8). This TAS only gets 4-C, but I suppose it's good enough. This strategy is essentially humanly impossible, however, as it requires manipulating Mario so that he never passes x-position 124. This allows Mario to stop at the edge of the second pipe and turn around.

Several in-level frames can be saved in 4-2 through the method demonstrated in HappyLee's video. It has been known for years, but I forgot about it when making this TAS. Still, this TAS enters the pipe with 4-H and enters World 8 with 8-7. KingOfJonnyBoy's 4-2 videos were based on this route.

Maru

Those mentioned framerules are difficult to save and require impeccable timing. I don't see those framerules being saved in runs any time soon.

Remember, these tricks are already somewhat deep into a run. That 4-2 clip comes after two flagpole glitches and one pipe clip, and saving the 8-1 framerule comes after all of that.

People have saved time over the normal speedrun route in 1-2 of SMB3 by sliding through the level (about half a second), although it's quite inconsistent, so it's never used. These tricks are just as difficult and inconsistent, but it occurs much later into a run. 1-2 is just the second level that's played in SMB3. For an SMB1 warps run, 4-2 is the fourth, and 8-1 is the fifth.

Both of these tricks have too many moving parts to be incorporated deep into attempts at the moment.