SI-generated
Speedrunning had one of those weeks where the word human started to feel like a technicality. A Mario player reached a split the community has treated as TAS territory, a machine-assisted Minecraft run turned cosmic rays into a routing tool, and the wider scene found itself revisiting an argument about who gets to reuse the community’s history. In between, records fell, personal bests landed, and somebody tried to explain why the correct way through World 1-2 is apparently to stop fighting the goombas.
The loudest story is Niftski’s latest Super Mario Bros. run, but it is not merely another improvement on a famous leaderboard. It is a glimpse at the narrowing gap between the best human play and the game’s theoretical limits. Elsewhere, Super Mario 64 kept proving that its competitive middle class is still terrifyingly fast, while new tools and old categories gave runners plenty of reasons to keep refreshing their feeds.
There was also a useful reminder that speedrunning is bigger than world records. A newly verified Resident Evil 4 Remake runner celebrated 13th place, viewers gathered around difficult ongoing attempts, and creators continued turning intimidating technical knowledge into guides and explanations. The week was about the ceiling, certainly, but it was just as much about everyone building the ladder.
Here are the biggest stories from the week.
Niftski Touches the TAS Wall — and Makes It Look Reachable
Some speedrun milestones arrive with a neat number attached. Others arrive as a collective intake of breath. Niftski’s 4:57.760 Super Mario Bros. run belongs firmly in the second category: it produced the first ever TAS tie on 8-4 during a live human run, a feat that sounds almost understated until you remember what that final level asks of a player. This is not simply fast movement performed cleanly. It is a chain of microscopic decisions, each one made at a speed where the game seems to be trying to turn your hands into a historical footnote.
The run’s context makes it even more remarkable. Niftski said this was literally the first attempt at the specific goal, and the community reaction settled on the same explanation: at this point, the biggest opponent may be nerves. The top discussion described a runner who has brought so many TAS-level 8-4s to the doorstep that the remaining challenge is not discovering whether the route works, but making the body believe it is allowed to finish.
There is a small asterisk, and it is exactly the kind of asterisk that makes speedrunning more interesting rather than less. The timer was reset partway through 8-4, so the video uses an edited timer and presents the accurate splits for comparison. The underlying play is still the story: the segment was substantially ahead of a conventional clean 8-4, helped by fast accelerations and movement that is usually filed under theoretical until a human demonstrates it.
That is why the comments quickly moved from “how fast is this?” to “what is left to tie?” The question is no longer whether humans can occasionally borrow a trick from the theoretical tool-assisted route. It is how much of the theory can be made repeatable under pressure. The answer is probably less than anyone would have guessed a few years ago, which is both inspiring and extremely rude to anyone trying to sleep.
Speedrun.com Reopens the Licensing Conversation
Not every important speedrunning story comes with a personal best or a thumbnail of Mario mid-jump. This week, Speedrun.com once again removed the Creative Commons licence from its terms, prompting a familiar wave of concern from runners, archivists, and people building tools around the scene. The immediate response was less “new controversy” than we have had this exact conversation before.
The 2021 version of the dispute ended with the licence restored after community backlash. This time, the site’s accompanying explanation reportedly framed the change as less consequential than the reaction suggested. That gap is where the anxiety lives. A leaderboard is not just a table of times: it is a community archive, a source for statistics, a foundation for projects that help runners discover and understand games. Changing the terms around that material can affect far more than the page a viewer sees after a run.
One highly upvoted response focused on a particularly important legal and practical point: material already published under a licence does not simply become unlicensed because the website changes its current terms. Other commenters were more blunt, greeting the announcement with the exhausted “here we go again” energy of a community that would rather be optimizing movement than rereading terms of use.
The discussion also highlighted the scene’s growing ecosystem. Independent projects such as therun.gg are not just competitors or side projects; they are evidence that runners want multiple ways to explore, mirror, analyze, and preserve the record. The healthiest outcome would be clarity that lets those projects operate responsibly. Speedrunning thrives on open knowledge, and it is difficult to build a durable archive when everyone has to wonder whether the rules changed overnight.
Cheese Finds Eight Seconds in a Game That Refuses to Sit Still
Super Mario 64’s 120-star category remains one of speedrunning’s great long-running dramas, and Cheese added another chapter with a 1:35:40 personal best. An eight-second improvement is huge at this level, particularly in a category where the route is long enough to offer countless opportunities for a run to quietly bleed away its potential. The result put Cheese fourth on the global leaderboard, another sign that the front of the category is not a museum display but an active, shifting contest.
The comments, naturally, noticed the outfit before the splits. A joke about the runner being “nekkid” collected more attention than several technically useful observations, followed by the inevitable theory that the look was inherited from Simply. That is not a criticism; it is a very SM64 way for a serious achievement to be greeted. The category can demand hours of obsessive practice, but its audience still reserves the right to spend the first thirty seconds discussing trousers.
Under the jokes was real recognition that Cheese has kept pace with a category whose standards have been brutal for years. The 120-star route is a test of consistency as much as execution: a runner must navigate a huge collection of movement-heavy stars, preserve focus through the middle, and still have enough left for the late-game sequence. Eight seconds is not one miracle. It is a stack of small victories that survived a very long run.
The TAS Classroom Is Back in Session
For anyone who wants to understand why Mario 64 can still look alien after all these years, Bismuth’s continuing TAS explanation series is doing the public service. Part two covers stars two through eight in a 45-minute lesson that treats tool-assisted speedrunning as something to be unpacked rather than worshipped from a distance. The appeal is not only seeing an impossible-looking route succeed; it is learning which inputs, camera choices, and bits of game logic make the impossible look inevitable.
The small comment section was enthusiastic in the way that matters: viewers were glad the series was continuing, one person declared that “we’re so back,” and another admitted they were there for the piano. That last detail is perfect. Technical explanation does not have to be dry, and a good presentation can make frame-perfect movement feel like both engineering and performance.
These explanations also give human runners a useful lens for the week’s headline. When Niftski approaches a TAS tie, the achievement lands harder if the audience understands what a TAS is actually optimizing. The tools do not merely press buttons faster; they search for lines that exploit the game’s exact physics and timing. Every human approximation of that line is a small act of translation between machine precision and muscle memory.
Minecraft Gets a 1:37 Assist from the Universe
There are random-seed runs, and then there is a Minecraft run that appears to have received assistance from common cosmic rays. A new random-seed speedrun finished in 1:37 using a radiation-assisted event, a sentence that sounds less like a leaderboard entry and more like the premise of a particularly anxious science-fiction short story. The mechanics are strange, the result is fast, and the explanation was impressive enough that commenters singled it out even while trying to work out what they had just witnessed.
The fascinating question is not only whether the run is quick. It is whether an accidental natural bit flip should count as part of a valid run. That is the kind of rules question speedrunning is uniquely good at producing: deeply technical, faintly absurd, and ultimately about what the category is trying to measure. If the category permits natural hardware behavior, is this an unusually lucky seed? If it depends on an external event that cannot be intentionally reproduced, does that cross a line?
One commenter compared the setup to forcing someone to play chess beside the Elephant’s Foot. The joke works because it captures the uncomfortable scale of the event, but it also points toward the real appeal. Speedrunning sits at the border between controlled execution and the chaos of a game engine. Most runners manipulate randomness through seeds, timing, and routing. This run asks what happens when the randomness manipulates the runner right back.
Whatever the eventual verdict, the run is valuable because it makes the rules visible. Categories have to decide what kinds of luck belong to the game, what kinds belong to the hardware, and what kinds make the result impossible to compare. That is not pedantry. It is how a community keeps a remarkable story from becoming an argument with no shared definition of “run.”
New Super Mario Bros. Keeps Moving at Cartoon Speed
The week also brought a new world record in New Super Mario Bros. Any%, clocking in at 22:08.817. It did not attract the same comment storm as the bigger stories, but that is often the strange fate of a record in a busy week: a major achievement can arrive quietly, framed by a title and a time, while the community is elsewhere debating cosmic radiation and the legal status of its archive.
There is something appealing about the contrast. The Mario 1 story is about shaving the last imaginable fractions from a foundational game; the DS-era platformer record is a reminder that “modern” Mario categories can still be compact, technical, and fiercely optimized. Twenty-two minutes is long enough for a run to develop a narrative and short enough for the runner to spend the entire thing moving as if the pause button personally offended them.
Records like this also reward the less glamorous work of the scene: route maintenance, setup practice, emulator and hardware knowledge, and the willingness to grind a category that may not dominate the front page every day. A world record does not need a 400-comment thread to matter. Sometimes the important thing is simply that somebody found the next line and proved it could be done.
The Runs We Are All Watching
A community thread asking which attempts people are currently following offered a useful snapshot of the spectator side of speedrunning. Niftski and Lekukie were the obvious headline picks, with the former’s Mario grind now carrying the tension of a live sporting event. Bubzia’s blindfolded 120-star attempts were recommended as surprisingly engrossing background viewing, which is a sentence that would have sounded like nonsense before speedrunning learned how to turn impossible challenges into comfort content.
Pokémon FireRed Round 2 was another recurring recommendation, helped by a $5,000 bounty with a deadline approaching. Bounties change the temperature of a category. They give runners a concrete reason to push now, bring more viewers into the search, and turn every near-miss into a possible piece of the solution. The thread also pointed toward other games and runners poised for breakthroughs, a reminder that the next headline is often being built in a category that is not currently trending.
That is the best argument for watching attempts rather than only watching records. A finished run tells you what happened; a live grind lets you see a route being negotiated in real time. You learn the reset points, the emotional weather, and the peculiar rhythm of a runner who may be one good sequence away from changing a leaderboard. The payoff is not guaranteed, which is precisely why the suspense works.
Thirteenth Place Is a Story Worth Celebrating
Among the records and near-records, one of the week’s most human posts came from a runner celebrating a newly verified 13th place in Resident Evil 4 Remake NG+ Any% on PlayStation 5. The category may not be enormous, and the game may no longer dominate conversation the way it did at launch, but that is exactly why the achievement matters. Speedrunning is not a popularity contest. A place on a smaller leaderboard still represents the hours required to learn the route, survive the execution checks, and submit a run that can withstand verification.
The runner only began in July and described the game as a running simulator punctuated by glitches, including a couple of techniques currently left unlearned because a mistake can kill the run. That is an unusually clear picture of how a speedrun grows: first you learn what the game wants, then you learn what the category demands, and eventually you decide which risks belong in your version of the route.
The replies were short but sincere: congratulations, encouragement, and a clear sense that “almost top ten” is a real milestone. It is easy to let the world-record end of the hobby distort our sense of progress. A verified 13th place is not a consolation prize. It is evidence that a new runner has joined the conversation and is close enough to the sharp end to start making plans.
That spirit appeared elsewhere too, in a Plants vs. Zombies runner compiling two years of learning into a guide and in a Super Mario Bros. player asking for help with the goombas in 1-2. The answer was gloriously simple on the surface — jump — but the follow-up advice got into early, small jumps and subtle left taps. That is speedrunning in miniature: a joke, a correction, and then a tiny technical detail that may save someone hundreds of resets.
A Week of Ceilings, Ladders, and Very Small Jumps
This week’s stories pulled in opposite directions. At the top, Niftski made the human-theory TAS gap look thin enough to cross, while Cheese reminded everyone that even a mature category can still surrender eight seconds. At the edges, a cosmic accident challenged the meaning of randomness, a licensing dispute challenged assumptions about ownership, and a new runner’s 13th place showed how much work happens away from the spotlight.
The common thread is that speedrunning keeps turning expertise into something shareable. Bismuth’s explanation makes TAS logic legible. Community comments turn a baffling trick into a route lesson. Guides collect years of trial and error so the next person can start a little closer to the action. Even the arguments about licensing come from the same instinct: the knowledge should remain usable by the people who created it and the people who will build on it.
There is always another frame to save, another category to revive, another runner quietly approaching the top ten. This week gave us records that looked machine-made, questions that were bigger than any one run, and enough tiny movement advice to keep the rest of us jumping at goombas for another seven days. See you next week.
Source: reddit.com