Speedrunning video games, the competitive field of playing through digital games as quickly as possible, has in recent years been elevated into something between a virtuosic form of fingers-and-thumbs athletics and a highly technical science. The best speedruns reduce epic games meant to take dozens of hours to single-digit minutes through a combination of exploiting glitch-enabled shortcuts and inhuman skill.

A little too inhuman, in some cases. Speedrunning, it turns out, is plagued with fake records set by cheaters who splice together video clips to falsify evidence or use rule-breaking software to gain unfair advantages. One speedrunner and hacker named Allan Cecil has made it his personal mission to catch them.

In a talk at the Defcon hacker conference in Las Vegas today, the details of which WIRED reviewed in advance, Cecil plans to present what he alleges is evidence that a speedrunning record for the 1996 PC game Diablo, which has stood for more than 15 years and holds a spot in the Guinness Book of World Records, was in fact the result of rule-breaking techniques that should disqualify it. If Cecil and the team of investigators who have worked with him over recent months succeed at tearing down that seemingly untouchable benchmark, it will be the third such high-profile speedrun that he’ll have helped to debunk, and the second in just the last year.

Cecil, who is better known in the gaming world by his handle dwangoAC, came into this strange role as a speedrun debunker from an equally niche hobby: He’s known as an expert practitioner of so-called “tool-assisted speedruns,” using emulator software to run a game in a controlled environment find the limits of what that game’s speedrun can be—an area of speedrunning that some purists once considered, itself, to be a kind of cheating. Cecil maintains instead that those tool-assisted speedruns, in which players meticulously rewind, replay, hone and perfect their runs frame by frame, can be its own valid form of competition, or even art.

Cecil says he arrived at his fixation with catching cheaters in part out of a determination to protect this lesser-known field of speedrunning from those who would surreptitiously use the same tools in deceptive ways, essentially turning tool-assisted speedruns into a kind of speedrun doping, rather than an honest avocation. “Making a tool-assisted speedrun is a transformative work of art that humans laboriously spend months on, or even years,” says Cecil. “But you definitely do not use tool-assisted speedrun techniques and then try to pretend that it was just human effort. And seeing people do that pisses me off.”

As a staff member at the tool-assisted speedrun website TASvideos.org and an organizer of many epic speedrunning feats—such as one that famously used coding glitches in the Zelda game Ocarina of Time to rewrite the game’s ending—Cecil has become a fixture of the speedrunning world. He’s also the creator of TASbot, a robot that connects to the controller ports of video game consoles to reproduce controller inputs, so that recorded speedruns can be replayed and verified on original gaming hardware, a spectacle that gamers love so much that livestreams featuring the robot have raised $1.5 million in donations to charitable causes, according to Cecil’s count.

In recent years, however, Cecil has taken his obsession with tool-assisted speedrunning in a different direction: He’s turned it into a means to catch the cheaters that threaten his pastime’s legitimacy. If he can show that even a polished tool-assisted speedrun in a certain game isn’t as fast as a purported human record—as he’s done in all three records he’s attempted to debunk—that demonstration can serve as the first step in suggesting that a record was likely falsified. He’s found, too, that the process of creating that tool-assisted run often produces new revelations about what’s possible—or impossible—in an unassisted human attempt.

Cecil’s latest record-busting effort may be the most controversial one he’s undertaken yet. At Defcon, he’ll present evidence that he alleges should disqualify the record of Maciej “groobo” Maselewski, a Polish speedrunner who holds the Guinness record for not only the fastest Diablo speedrun but also the fastest role-playing game speedrun of any kind. Maselewski’s Diablo run, 3 minutes and 12 seconds long, has withstood all challengers since 2009.

Cecil says his suspicions that Maselewski had broken speedrunning rules were first triggered when he and another speedrunner, Matthew “funkmastermp” Petroff, set out to make a tool-assisted speedrun for Diablo in January. They quickly came to the conclusion they’d never match Maselewski’s 3-minute-12-second time, no matter how much they perfected their run or how lucky they got in the game’s randomized dungeon layouts. That led them to assemble a team of investigators who ultimately found what they believe to be a long list of inconsistencies in software versions and items, missing frames, and other signs of potential tampering in the video of Maselewski’s run, all of which they’ve assembled in a detailed document posted to Cecil’s website. “No one could get anywhere close to that time. Now we know why,” Cecil says. “The answer seemed to be that groobo had cheated in many, many ways.”

Maselewski, when WIRED reached him for a response, immediately denied any such foul play. In an email, he pointed to the fact that his run was always understood to be “segmented”—essentially edited together, level by level, a commonly accepted category of speedrun. “It was never passed off as anything else,” Maselewski writes. “Hearing that there has been a team of researchers working on this is wild.” In a subsequent text exchange with Cecil and his collaborators that Cecil shared with WIRED, Maselewski described the attempt to debunk his record as a “witch hunt.”

Maselewski’s simple explanation that the speedrun was segmented, however, isn’t enough, Cecil argues. He claims that some dungeon layouts in Maselewski’s run couldn’t be generated even in a single segment of a run without altering the game’s data, and in one case, a crucial item conveniently appears that defies the game’s logic—Naj’s Puzzler, a staff that allows the player to teleport across distances—and alleges a piece of performance-enhancing software known as a “trainer” must have been used, all of which could disqualify Maselewski’s run. (Maselewski did not immediately respond to WIRED’s follow-up question specifically about Cecil’s allegation that he had used a trainer.)

The night before Cecil’s Defcon talk, Maselewski wrote in a final email to WIRED that he believes those alleging that he cheated are using faulty tools with an incomplete picture of Diablo‘s complexities. “Dwango is out to tell a story. Did I cheat? No,” Maselewski writes. “But what is true or not does not matter at this point, because the wonder of exploration has already overstayed its welcome for a small group of people, and the script has already been written.”

When WIRED reached out to the Guinness Book of World Records to ask if it would take down Maselewski’s record, a spokesperson responded noncommittally that “we value any feedback on our record titles and are committed to maintaining the highest standards of accuracy.” An administrator for Speed Demos Archive or SDA, another speedrun record-keeping website where Maselewski holds a similar Diablo record, seemed to be more persuaded by Cecil’s evidence. That administrator, who goes by the handle “ktwo” and asked that WIRED not include their real name, says that SDA hasn’t officially reached a verdict and is still waiting to hear Maselewski’s explanation.

Things are not looking good for groobo, however. “To be clear, we have made a preliminary decision, based on the available information,” ktwo writes “The staff agrees that the analysis raises questions about the validity of the run that need to be addressed, or else the run will be unpublished from SDA. The admin team is currently discussing these questions with the runner. Once that discussion has concluded, a final decision will be made.”

Cecil’s involvement in investigating gaming records began in 2017, when the speedrunner Eric “Omnigamer” Koziel, who was writing a book about speedrunning, began re-examining a record set by Todd Rogers for the Atari 2600 racing game Dragster. Rogers’ record time, 5.51 seconds, had persisted for a remarkable 35 years. But when Koziel reverse engineered Dragster’s code to try to understand how Rogers had achieved that time, he found that tricks Rogers said he’d used—such as starting the game in second gear—wouldn’t have provided the advantage Rogers claimed.

“The goal was never to point to someone and say, ‘Hey, they’re cheating,’” says Koziel. “It was to try to find the truth.”

Cecil, who knew Koziel from the speedrun community, offered to help develop a tool-assisted speedrun they could replay via TASbot on a real Atari 2600 to show that, even on that original hardware, Rogers’ record was impossible. They found that TASbot’s theoretically perfect performance was 5.57 seconds, slower than Rogers’ alleged time. Despite Rogers’ objections, his three-and-a-half-decade-old record was erased from the annals of the gaming records keeper Twin Galaxies—along with all his other records on the site—and Guinness stripped his world record for “longest-standing video game record.”

“Although I disagree with their decision, I must applaud them for their strong stance on the matter of cheating,” Rogers wrote in a lengthy public Facebook post responding to the Twin Galaxies decision.

After a seven year hiatus in which he was mostly focused on TASbot projects, Cecil found himself involved in investigating another legendary speedrun earlier this year when a group of gamers made it their goal to beat every level of Super Mario Maker. That Wii U game, released in 2015, allowed users to upload their own levels of the game for others to play—if they could prove that the level was beatable by recording a video of themselves finishing it. Yet one such level, called “Trimming the Herbs,” appeared to be impossible. For years, no one but its creator had been able to complete it.

In an effort to help this group of Super Mario Maker-obsessed gamers, Cecil offered to develop a tool-assisted speedrun for the level. He found that it was nearly impossible to clear it reliably, in part due to variances in the Bluetooth communications between the Wii U and its controllers. In that case, the level’s creator came forward in the midst of the investigation and confessed he’d beaten the level by altering the internal hardware of his Wii U gamepad—a trick he intended as a kind of friendly prank but had never publicly explained.

In his new attempt to debunk Maselewski’s Diablo record, Cecil doesn’t expect any such amicable agreement on the facts—or a confession. But he’s confident that he and the researchers who’ve worked with him can remove Maselewski’s record and make room for speedrunners to approach the game again. To Cecil’s surprise, they found just in recent days that they could, in fact, beat Maselewski’s record with a tool-assisted speedrun—thanks to new Diablo techniques they found in their investigation—finishing the game in 2 minutes and 45 seconds without using any of his alleged rule-breaking alterations. Meanwhile, Cecil says Diablo speedrunner “xavier_sb” has completed a run of the game in just over 4 minutes and 40 seconds, which would stand as the new record if Maselewski’s is expunged.

Cecil argues that shows, already, how the “chilling effect” of what he alleges to be an impossible record is lifting. “People had just stopped trying, because there wasn’t any point,” Cecil says. Now the Diablo speedrunning race is on, once again.

Cecil says he hopes that his work to keep speedrunners honest will not only help traditional speedrunning to thrive, but fuel the tool-assisted kind that he has helped pioneer, too. The key, he maintains, will be drawing a clear line between those who use software tools to play games with inhuman precision as an honest art form, and those who use them for deception. “Both the jerk and the jester bend the established rules. We despise the vileness of the jerk and delight in the shenanigans of the jester,” Cecil says. “But one thing has held true: No one likes a cheat.”