How Tool-Assisted Speedrunning Reveals The Inner-Life Of Video Games

If you watched this year’s Awesome Games Done Quick, you’re already familiar with the Olympian feats of endurance and skill that players exhibit as they race through your favorite games and make you realize that, no, you aren’t nearly as good at them as you might have thought. But there’s another side to the coin that isn’t nearly as popular as traditional speedrunning, a niche community of players prying open games and doing the unimaginable. Tool-assisted speedrunning (‘TAS’) might seem like simple cheating, but it’s so much more than that. It’s an exploration of the inner-life of video games, and it takes teams of expert players and programmers thousands of hours to do.

For many of us, a video game ends when the credits roll or when every challenge has been mastered—if we even make it that far. But for Scott “pannenkoek2012” Buchanan, Super Mario 64 has no end. It’s a game that he has mastered both inside and out, and his knowledge of its inner clockwork borders on academic. By using emulators and memory hacking software, Scott is able to play Mario at a level that is impossible by traditional means. He can control Mario frame by frame, maneuvering him in ways even the average speed runner only dreams of doing. His staggering understanding of the game’s programming and the glitches it spawns allows him to combine both to fascinating effect.

“I’ve been playing Mario since before I can remember,” Scott told me one afternoon as we chatted over Skype. As a child, it was the first game he ever played. But unlike most of us, once Scott had finally mastered every challenge Super Mario 64 offered, he wasn’t ready to move on. Using sm64.com, Scott’s “bible” for the game, he began seeking greater challenges, like collecting every coin on every level. Through the website, he even discovered there were three different levels that, with careful use of a glitch, a player could potentially gather infinite coins. But being just a kid wrestling with the N64’s awkward controller, Scott found he lacked the dexterity for these greater challenges and eventually drifted away from the game.

It wasn’t until several years later, as a teenager, that Scott rediscovered his love of Super Mario 64 thanks to an emulator he downloaded onto his computer. Adamant that he would now beat the challenge his childhood self had failed to master, Scott discovered something interesting: It wasn’t just three levels that Mario could gather infinite coins on but many more. Excited by this new discovery, Scott made his first YouTube video to share his findings with the world and took his first steps into the community of tool-assisted play.

Meet The TAS Team

Today, Scott is one of the more popular tool-assisted runners in the entire community. He works with a team of 12 other players who have transformed Super Mario 64 from a game into a science. But not content to be just scribes, they use their masterful knowledge of these glitches to do incredible things. In 2014, for example, Scott collected a coin that players had known about for 12 years but been unable to reach by exploiting geometry in the world.

Scott’s main focus now is the ”A’ button challenge’, a bid to figure out the optimal way of beating Super Mario 64 both by collecting all 120 stars (called a 100% run) or going straight for the finish line (called an any% run). Thanks to Scott and his peers, the ‘A’ button challenge has been reduced from 118 presses down to 32 for the 100% run. He’s even working on a “no button challenge” to beat the game using only the joystick.

A graph charting the decreases in 'A' presses and the glitches discovered that made that possible.

This incredible reduction of presses is thanks to Scott and the Facebook group of 12 other Mario 64 ‘TASers’ he collaborates with. “Each one of us has a really good strength that no one else has, so it’s really helpful that we’re all together and can communicate so easily,” he said. When I asked Scott what he felt he brought to the group, he was quick to answer: “I’m creative in coming up with new strategies and I have a lot of patience. If something doesn’t work I’ll keep trying it and keep thinking about new possibilities.” That patience goes well beyond running into walls when trying to figure out a new method. Scott’s tool-assisted runs can frequently take him up to 12 hours to complete a single movement—often just to save a single ‘A’ press.

Tyler Kehne is another influential member of the inner circle of Mario TASers. When Scott’s videos were first gaining popularity, it was common for viewers to suggest solutions in the comments, but the vast majority of them were coming from people not nearly as educated as Scott. That was until Tyler started reaching out and blew Scott away with his knowledge of the game. “He started showing me stuff and I was like, whoa, this is a guy to be reckoned with. He showed me things I had never thought of,” Scott said.

Tyler uses his skills as a computer programmer to unravel Super Mario 64’s more esoteric secrets. He was the person who discovered the complicated trick of ‘parallel universes’, a term he coined after discovering a way to desync Mario’s actual position on a map and the position the game uses to check if he is standing on solid ground.

Parallel universes were a massive breakthrough for the ‘A’ button challenge because, paired with another glitch called ‘hyperspeed walking’, it allowed Mario to gather enough speed to essentially break free from the map entirely, even though his ground-check position still remained behind. This trick places Mario on an invisible duplicate of the map, since the game thinks Mario is technically on solid ground when he is in fact way out of bounds. These duplicates extend in a grid an infinite number of times. With enough built up speed a TASer can move Mario around a level by causing him to break through the geometry and travel to new parallel universes, thus circumventing obstacles that would otherwise require an ‘A’ press. If it sounds complicated, that’s because it is.

What Tools Are Used For Speedrunning

Such complicated methods often require complicated calculations, something that Scott excels at. During our chat, he showed me his workbook containing over 20 different spreadsheets of all of his research. Everything from the flight paths of certain enemies to calculations of how much speed Mario needs to perform a parallel universe movement are meticulously detailed and graphed.

Calculations and graphs Scott created to chart the flight path of an enemy.

The values for his calculations are gleaned through the use of memory hacking software like Cheat Engine, which can peek into specific addresses in RAM that a program assigns and read back whatever is stored there. Variables like Mario’s coordinates or the vertices of polygons can be determined, but doing so requires first knowing where to look.

An important distinction to make is that everything Scott and his peers accomplish is theoretically possible using an unassisted version of the game. Though they certainly have the tools to hack Super Mario 64 ROMs or pry open the values stored in specific RAM addresses and change them as they please, these methods are only used to study and understand the game and not in the actual runs used to complete challenges.

One member of the group is even working on a modified controller that can interface with an N64 console and input the same sequences of button presses used on emulators to verify that what is done on an emulator is possible on the native hardware. This was most recently done to verify a method Scott had devised by cloning several goombas to make a ladder to avoid an ‘A’ press. It was a process that, due to the limitations of the N64’s hardware, he was worried would crash the console if it ran out of memory, but was relieved to discover it didn’t.

When it comes to actually performing the run, all Scott uses are the tools provided by the emulator, a controller plugin to offer greater precision, and the memory hacking software to see any values he might need in order to perform his complex maneuvers. He also uses a wired Xbox 360 controller to play the game because the extra buttons allow him to assign necessary emulator functions that he needs easy access to.

In a demonstration, Scott showed me how to exploit a ‘misalignment’ to allow Mario to snap up onto a higher ledge without using the ‘A’ button. By first standing on the ledge, Scott was able to determine the specific vertice he needed Mario to jump into to cause him to teleport upwards. Then, using save states to retry and correct failed attempts and Xbox 360 buttons to move the game forward frame by frame, he had Mario dive (which doesn’t require the ‘A’ button) toward the vertice several times. Using the controller plugin to correct his movement by a fraction of degrees, Scott was able to time the jump perfectly to collide with the level geometry in such a way that he would be pushed up onto the ledge instead of away from it. In total, the whole process took minutes to complete a move that, in real time, would have been seconds.

On page two, how glitches are found in the first place.

22 Comments

  1. Urthman says:

    The video behind your “that’s because it is” link is fascinating. He explains everything so well, and is clearly having a great time blowing your mind.

    link to youtu.be

    I especially love the bit, which feels like something out of a science fiction novel like Tau Zero, where he essentially says, Unfortunately we can’t just teleport to the mirror galaxy 27 billion light years away, because the intermediate stutter steps between the galaxies would kill us. But we can step beyond the mirror galaxy all the way to the next mirror galaxy. Four times. Four galaxies for each of the dozen-meter hops we want to take in our world, making sure to arrange our path so that the final set of four-galaxy jumps lands us back in our own galaxy.

    (Also, the planets in the mirror galaxies have floors, but no walls, objects, or creatures. And if you open your eyes, you die.)

    • April March says:

      I randomly saw one of his videos yesterday, and I agree – he’s just so nice to people who are antagonizing him, and he explains what he’s done in an understandable way without ever coming across as smug. We ought to cherish him.

  2. Eschwen says:

    As I understand it, there’s a lot of cross-polination between the tool-assisted groups and the pure groups. Watching AGDQ this year, it was mentioned a few times that certain glitches or techniques used by games were uncovered by the tool-assisted teams, which then allowed the pure guys to improve their times or approaches to certain parts of games.

    For example, I believe it was mentioned during the last speedrun of Final Fantasy 4 that the frame on which you exit the main menu becomes the seed that determines monster encounters, which was discovered by some tool-assisted runners. This is now used in combination with step charts to plot optimal minimal-encounter routes through the game for pure speedrunning. So it would be pretty silly for people who enjoy pure speedrunning to scoff at the tool-assisted approach, as both are integral to enjoyment we get from the end result!

  3. KDR_11k says:

    Yeah, TASes often get dismissed far too easily, people think that because the tools remove the reflex challenge it’s easy to make a TAS. With great power come great expectations and the level of perfection that’s expected of a TAS these days is extremely high. You have to use every last frame saving trick that’s known and often even ones that aren’t. RNGs must be manipulated (old consoles don’t have a proper source of randomness and thus tend to seed their RNGs with controller inputs), lag frames avoided and of course it goes without saying that every corner must be taken perfectly.

    Oh and after all that? The goal of a TAS is to entertain, if it’s judged to be boring it still gets rejected.

    It takes several months to do a TAS of a game like Mega Man. Often a source of imperfection in TASes is that new techniques get discovered while the run is already months in development and they can’t be integrated without desyncing.

    BTW, while it seems that speedrunners will take any version (usually JP for shorter texts) and configuration of a game to make it faster a TAS must use the US version (if available, exceptions only when other versions allow major new tricks), run on the highest difficulty available and start with no data saved on the cartridge. Exceptions to the last part are only for explicit NG+ runs and often those NG+ saves are generated by another TAS. Of course it’s important to prove that the run does nothing that can’t be done on a real console.

  4. snoof378 says:

    Thank you, Steven, for a really interesting and well-written article. I enjoyed reading about how these things are discovered and I am in awe of what Scott and co. are able to achieve within the technical limitations of the unaltered game. They’ve created their own challenges more befitting of their knowledge and creativity than the game itself, turning it from a platformer to a puzzle of physics and mathematics – awesome!

  5. Arglebargle says:

    What really did my head in was discovering that the Mario games were originally envisioned and designed to star Popeye, til some late legal IP crap intervened.

    Never been able to look at it the same since.

  6. guygodbois00 says:

    Spider charts. Because nobody likes charts and even less the arachnophobia!?

  7. DevilishEggs says:

    I’ll never make fun of console gaming ever again.

  8. Wowbagger says:

    Dustforce is another game where TAS is in evidence, it’s strangely hypnotic to see someone working on a run in that way.

  9. April March says:

    My own thoughts echo this: link to twitter.com

  10. Johnny Go-Time says:

    Fantastic article! I thought I had zero interest in the topic, but this was great.