The Machine Is Right and Nobody Believes It

On July 2, in a round-of-32 match at this World Cup, Croatia scored what looked like a late equalizer against Portugal. The players peeled away to celebrate. The stadium did what stadiums do. And then the goal came off the board — because the tournament’s semi-automated offside system had detected a touch by Igor Matanović on the ball’s way through, a touch that does not appear on any broadcast replay, at any speed, from any angle.

The next day, FIFA affirmed the call. VAR and the semi-automated system had functioned exactly as designed; the touch was real; the goal was correctly disallowed. Croatian fans, and a healthy share of neutral ones, reached the opposite verdict: robbed.

Here is the uncomfortable part. Both sides are almost certainly right. The machine detected something real, and the fans were shown nothing they could check. Automated officiating’s problem in 2026 is not accuracy — by any honest reading of the evidence, the machines are better at this than we are. The problem is legitimacy, and the institutions running the machines are failing at the one thing that has ever bought legitimacy: showing the governed the evidence.

I have watched every match of this tournament, the way I have watched every match of every World Cup since 1990. I want the calls to be correct. I carry no nostalgia for the linesman’s guess. And when the Croatia decision landed — on the broadcast, in the group chats, across every second screen — what I watched was not a technology failing. It was an institution declining to make its case.

The Evidence You Cannot See

The system that made the call is genuinely remarkable. FIFA’s advanced semi-automated offside runs on skeletal tracking of every player on the pitch, a 500Hz inertial sensor inside the adidas Trionda match ball, and per-player 3D avatars built from full-body scans of every squad in the tournament. When the system detects a clear offside, the alert goes straight to the on-field officials’ earpieces. It does not even pass through the VAR relay anymore. The design goal is obvious: kill the four-minute wait, deliver certainty at the speed of play.

It has not quite gotten there, even for the people wearing the earpiece. The Guardian interviewed World Cup assistant referee Micheal Barwegen in June, and he described the experience from inside the loop: “The computer has to think, and it’s super fast, but [on the field] it feels like forever.” The humans, it turns out, still out-call the machine on speed. The machine wins on certainty. Everyone on the pitch is now waiting for the same verdict the rest of us are.

And how certain is it, exactly? Here is a telling detail. What FIFA has disclosed is a threshold: the upgraded system now sends its audio alert when a player is more than ten centimeters offside, down from fifty at the last World Cup. What it has not published is an error bar — how confident the system is in any given ten-centimeter judgment, how often it might be wrong, and by how much. The precision of the most consequential officiating instrument in world football is, as of this writing, an undisclosed number.

Which brings us back to the Matanović touch. It is below the threshold of human perception. The replay cannot show it because the cameras did not resolve it; the only witness is a sensor inside the ball, sampling five hundred times a second, legible exclusively to the people who operate it. The call may be perfect. It is also, to every human being who does not work for FIFA, unfalsifiable.

That is a new condition in sports.

Showing the Work Only When It Flatters

FIFA does release evidence — sometimes. The tournament has produced offside animations for plenty of decisions, rendered avatars frozen at the moment of the pass, the kind of exhibit that settles arguments — even for calls nobody disputed. But on June 13, when an offside check stood between Switzerland and a penalty against Qatar, the avatar animation never came. FIFA blamed a “brief technical outage,” and the evidence it finally produced, four and a half hours after the call, was a pair of old-fashioned hand-drawn VAR lines the BBC judged unconvincing. Gary Neville was calling the withholding “like a dictatorship” on air before the statement even landed. The avatar exhibit for that call has never surfaced. Animations for the routine cases, a four-hour vacuum for the contested one.

It scarcely matters that the cause was a malfunction rather than a decision; the audience only sees the pattern. Evidence that appears instantly for the routine calls and hours late for the contested ones teaches the audience exactly what the disclosures are for. A court that publishes its reasoning only when the ruling is easy is not somewhat legitimate. It is performing legitimacy, and everyone watching can tell.

By early July the discourse had found its register. Forbes asked “Has Technology Ruined The World Cup?” the day after the Croatia decision; the BBC’s The Inquiry had devoted an episode in late June to “Is technology ruining sport?” The framing is wrong in both cases — the technology is working, probably better than any officiating instrument in the history of the game. But when mainstream outlets reach for the word ruined about a system with a near-flawless accuracy record, that tells you something important about where the deficit actually is. It is not in the sensors.

It is worth saying that the machine giveth too. On June 19, the United States had a Freeman goal awarded against Australia after the system’s automatic review overturned an on-field offside flag. Nobody demanded the sensor data. Nobody ever does when the call goes their way — which is precisely why the disclosure policy cannot be improvised match by match. Legitimacy is never tested by the calls in your favor. It is built, in advance, for the day the machine rules against you on evidence you cannot see.

The League That Published Its Error Bar

Now look at baseball, which is running the same experiment with the opposite disclosure posture.

Since Opening Day, MLB has been live with the ABS challenge system: Hawk-Eye tracking underneath, two challenges per team, retained if successful. Only the pitcher, catcher, or batter can challenge, with a tap of the cap inside about two seconds — no dugout committees, no replay huddles. By mid-April, per CBS Sports, teams had thrown 1,050 challenges at the machine, at a rate of about 4 per game and covering roughly 1.4 percent of pitches. The success rate: 54 percent.

Two design choices separate MLB’s system from FIFA’s, and neither is about accuracy. First, MLB published its machine’s fallibility. The league’s stated precision — figures given to The Athletic — is 95 percent confidence within 0.39 inches, 99 percent within 0.48. Read that again: the machine has an error bar wider than some of the margins it overturns. MLB printed that fact anyway. Second, the whole dataset is public. Every challenge of the season sits on a live dashboard at Baseball Savant, auditable by anyone with a browser, and the animation of every challenged pitch plays on the broadcast and the ballpark videoboard within seconds of the cap-tap. The evidence is not released on appeal. It is the product.

Has radical disclosure destroyed faith in the system? The opposite, as far as anyone can measure. One survey of two thousand Americans in May found 21 percent saying ABS will make them watch more baseball, and 54 percent saying it moves the needle not at all — automation as furniture, not crisis. Challenge skill has become a scouted attribute: Cubs catcher Carson Kelly went 21-4 on challenges through mid-May, worth 2.3 runs above expectation per Baseball Savant. Fans argue about whether their catcher challenges well, the way they argue about whether he frames well. The machine got absorbed into the sport’s argument culture instead of ending it.

The most interesting finding in the half-season data is stranger still: the machine changed the game without making a call. Players told the AP in April that the effective zone has gotten tighter — umpires now get instant feedback on every challenged call, and the borderline strike is disappearing; hitters have adjusted, and walks through the season’s first month ran at 9.8 percent of plate appearances, a rate the league had not seen since 1950 — a jump the AP called massive even adjusted for April’s usual walk inflation. The machine’s behavioral shadow is bigger than its interventions. That is a real cost, and worth its own essay. But note that we know about it — players discussed it openly, reporters quantified it, the league’s own public data made the analysis possible. In an open system, even the second-order problems get litigated in daylight.

Baseball shows its work. Football files the work in a drawer and issues a statement.

The Control Experiment

If you suspect the answer is simply to keep humans in the chair, tennis ran that experiment this season, and it is worth watching closely, because tennis ran both arms of the trial at once.

The French Open is the last Grand Slam still using human line judges. Amélie Mauresmo defended the choice on the grounds that “the machine is not 100 percent reliable.” She is right — no machine is, as MLB’s published error bar cheerfully concedes. And then her tournament produced the worst line-call scandal of the tennis season, a missed call in Ruud–Fonseca on May 31 that had Jim Courier saying on air that electronic line calling “makes far fewer mistakes than humans.” The clay-court distrust runs the other way too: Elena Rybakina, after a no-mark call went against her in Madrid this spring, said she “won’t trust it at all.” And the worst automation incident at Wimbledon remains last July’s Pavlyuchenkova–Kartal game, in which the electronic system missed a call because a human had switched it off — the club apologized. The failure modes have migrated from judgment to operations, but they have not disappeared.

So: the humans miss calls, the machines get switched off, the players trust neither, and the tournament that kept the humans got punished for it hardest. Neither arm of the experiment produced trust, because trust was never a function of who makes the call. It is a function of whether the governed can check the call. Tennis’s electronic calls at least leave a mark on a screen. The French Open’s human calls leave a mark on the clay — which is exactly why players stand and point at it. Everyone in that sport understands, instinctively, that the argument is settled by shared evidence. The only unforgivable state is the call with no mark at all.

Fifty Years in the Bar

For half a century, the story of officiating evidence was a story of democratization.

Instant replay’s deal with the audience was radical when it was new and invisible by the time I was watching cricket and football as a kid in India: everyone in the bar saw what the referee saw. Slow motion, then more angles, then super-slow-motion, then every phone in the stadium. The entire argument culture of modern sports — talk radio, the pub table, the group chat, first-take television — is built on shared access to the same frames. The referee had authority, but the evidence was common property. Even the VAR era, for all its agonies, kept the deal intact. The freeze-frame with the lines drawn on it was mockable, contestable, endlessly relitigated — but it was visible. You could hate the lines and still see them.

Cricket, the sport I grew up arguing about, understood this better than anyone and turned it into theater. When a review goes upstairs, the evidence plays on the giant screen for the entire ground — the ball-tracking arc, the audio spike on the edge, each stage of the verdict revealed to sixty thousand people at once, who gasp and groan at every frame like a jury that has been handed the exhibits. The decision review is one of the most-watched moments in the modern game precisely because the crowd is inside it. Nobody in that stadium is asked to trust a sensor they cannot see. The evidence is the show. That is what a tracking system looks like when the institution running it decides that the audience is a party to the proceedings rather than a security risk.

Sub-perceptual tracking quietly reverses the whole fifty-year arc. The decisive fact in Croatia–Portugal is not in any video. There is no frame to freeze. The evidence has moved from the screen — where everyone could see it — into a data stream only a priesthood can read. Officiating evidence, democratized for two generations, is being re-aristocratized in a single tournament cycle. You are no longer shown the proof. You are asked to trust that it exists.

Verifiability, not accuracy, is what legitimacy runs on.

The tempting frame here is man versus machine, and I have written before about the danger of binaries — this one obscures more than it reveals. Because the men signed up. NFL officials ratified a seven-year labor deal by a 116-4 vote this spring, months after a season in which the league ran virtual first-down measurement league-wide. MLB’s umpires accepted ABS inside a routine collective bargaining agreement. Barwegen, the World Cup assistant referee, says the system makes his job easier in some ways. The officiating labor war that a decade of think pieces predicted simply never happened. The real division in automated officiating is not man versus machine. It is open versus closed.

And closed is about to scale. Adam Silver said in May that NBA out-of-bounds calls “will be done by an AI-automated system with cameras lined around the court.” Every major league is walking toward the same architecture — tracking layers that adjudicate below the threshold of what a camera can show a fan. Whether the governed get to see the evidence is being settled right now, league by league, mostly by default, one press release at a time.

Sunday

This essay publishes on a Friday. The World Cup final is two days later, on Sunday, at MetLife Stadium. Somewhere in that match — a tight offside, a disallowed goal, a penalty awarded off a touch nobody in the stadium perceived — there is a decent chance the machine will speak. It will very probably be right. The question FIFA has two days to think about is whether anyone will believe it.

The fix does not require a task force. It is a policy, and it fits in a sentence: publish the evidence, every time. Every semi-automated decision releases its full case at the moment of the call — the animation, the sensor reading, the margin, the confidence interval — as a matter of routine rather than discretion. Not only when the exhibit is flattering. Not withheld when the margin is embarrassing. Not on appeal. The institutional instinct says that showing uncertainty invites doubt. Baseball just spent half a season demonstrating the opposite: MLB published an error bar wider than its overturned margins, put every decision on a public dashboard, and fans metabolized it into the ordinary texture of the sport within weeks. The disclosure absorbed the doubt. The drawer manufactures it.

Sports adopted these systems because the calls matter — because a World Cup turns on a fingertip, and getting the fingertip right is worth building a 500-samples-a-second sensor into the ball. But the call was never the product. The shared conviction that the call was fair is the product, and that conviction has only ever been built one way, in courtrooms and elections and penalty boxes alike: show the governed the evidence.

On Sunday, the machine will almost certainly be right. Whether anybody believes it is a decision FIFA still gets to make in advance.

Make the right call. Show us.