World Cup Round of 16: Portugal vs Spain, the Hosts Against the Machine, and UFC Week Begins

The last time our whole panel agreed on a knockout game, Norway made all four of them wrong. So tonight comes with a warning label. The panel splits on Portugal and Spain, and the one call it does agree on is the one our American readers (and this writer) are going to hate.

Yesterday, graded

Start with the receipts, because Sunday was humbling. Norway beat Brazil 2-1, and every forecaster we run had Brazil: the analyst at 72%, the diehard at 72%, the scout at 65%, the sharp at 73%. There is no way to dress that up. Four models, four misses, on the same game. When the whole panel clusters like that and loses, the useful takeaway is not "models are broken," it is that Norway keeps living in the part of the probability distribution everyone assigns and nobody expects. We log every one of these on the accuracy page, wins and stinkers alike.

The Azteca game went the other way. England survived Mexico 3-2 with ten men for the last half hour, and the majority of the panel had England at 65%. The scout was the lone dissenter on Mexico at 52%, a call we flagged in Saturday's post as the interesting one to watch. It was interesting. It was also wrong. The scout is now the only persona carrying two solo knockout losses, and tonight, as you are about to see, he is alone on both boards again.

The game itself was fantastic. No level of modeling, prediction, or AI could have told us we were in for absolute cinema but absolute cinema is what we got.

Tonight: the Iberian derby

Portugal vs Spain is the kind of game the panel was built for, because the models genuinely disagree. The analyst has Spain advancing at 68%, built on the attack gap (2.01 expected goals per game to Portugal's 1.70) and the fact that Spain carries 19.1% title odds to Portugal's 2.0%. The sharp is even colder at 70% Spain. The diehard, who weights recent fire and cup pedigree more, only gets to 57%. And the scout picks Portugal outright at 54%, alone on the board for the third time in three days.

That is the widest four-model spread we have had in a knockout game: 24 points between the scout and the sharp on the same match. Somebody's process is going to look very good at midnight.

The hosts against the machine

We are an American publication. The office is cheering for the USA tonight, unapologetically. The problem is that we also publish a forecaster that does not care how we feel, and the sharp has Belgium at 63%, with the analyst even lower on the hosts at 72% Belgium and only the scout willing to take the USA at 52%.

Here is why the sharp's number stings more than the others. We just ran its knockout calibration, and it is 18 for 20 on advance calls this tournament, 90% right while claiming an average confidence of just 69%. It runs under-confident: its "lean" calls (50 to 65% stated) have hit 100% so far, its favourite-band calls hit 85%. When this model says 63% Belgium, history says the true number has been higher than that, not lower.

The two games it missed? Germany vs Paraguay and Brazil vs Norway. The two biggest upsets of the tournament, the exact kind of night every USA fan is praying for. So the honest framing is this: the machine that is almost never wrong says the hosts probably lose, and the only two times it has been wrong all month looked exactly like the miracle Americans want. Follow it live on the World Cup hub.

Norway is real

One more thing from Sunday before we move on. Norway's win was not a fluke result our models grudgingly absorbed; the title odds moved hard. Norway now sits fourth in our tournament simulation at 8.5%, ahead of Morocco (8.2%) and England (7.7%), behind only Argentina (24.5%), France (20.1%) and Spain (19.1%). A team that started the knockouts as bracket filler is now, by the numbers, likelier to lift the trophy than England. That is two giant-killings deep, and the sim has stopped treating them like a guest.

UFC 329 week: one fight a day

UFC 329 is Saturday, and rather than cram fourteen fights into one preview we are covering one fight per day all week, working up the card to McGregor and Holloway. Today: King Green vs Terrance McKinney at lightweight, because no fight on the card is easier to state as a math problem.

Terrance McKinney's average UFC fight lasts 2 minutes and 16 seconds. That is not a typo. Across 13 fights in our log, his median night is over before the first commercial break: eight first-round finishes landed, three suffered, and his last five fights ended at 0:24, 2:30, 0:55, 2:01 and 0:37, every single one in round one. Our new per-round data makes the pattern brutal. In round ones, McKinney averages 17 significant strikes, a knockdown every fourth fight and a minute of control time. In the two fights where he saw round two, he landed 4 strikes total and controlled his opponent for five seconds. His opponent-adjusted grappling offense rates elite (+0.77) while his grappling defense rates among the worst we track (-0.78). He is a firework, in both directions.

King Green is the counter-case. He is 39 years old, 35-17-1 over a career that started when McKinney was in middle school, and his average fight runs 9 minutes 12 seconds. His output climbs as fights age: 27 significant strikes in round one, 40 in round two, 51 in round three, and his knockdown rate actually peaks in the third. He is on a three-fight win streak with a fun footnote: one of his last losses came to Paddy Pimblett, who co-mains this same card.

The ratings call it nearly even, Elo 1742 to 1734 for Green, though Glicko likes Green wider (1711 to 1641). The league context splits the difference: younger fighters win 61.5% at lightweight (McKinney is 31), while the rating favorite wins 58% (that is Green, barely). Green has also been finished in the first round four times since 2020, so his durability early is not a given.

The read: this fight has a clock on it. Everything McKinney has ever done well happens before the 5:00 mark of round one, and everything Green does well compounds after it. Watch the full matchup page for the tale of the tape and the round-by-round league trends. Descriptive, not a prediction, and two round-twos is a tiny sample, but rarely does a fight's entire question fit in one sentence: can Green survive five minutes?

The MLB board: one unanimous pick and a six-run gap

The scout report flagged four things worth your evening on an eight-game slate.

The early game is the strange one: Phillies at Royals drew total alignment, the base model at 58.6% Philadelphia, both personas on Philadelphia and the under, and all six AI-field models clustered around 62% with a spread of less than two points. Vega's case: "Sánchez's 2.31 FIP and 24% K-BB% towers over Cameron's 3.83 FIP; PHI's bullpen edge (3.60 vs 5.18) seals it." The buried detail is that both starters are left-handed, a genuinely rare matchup shape, and Edmundo Sosa happens to hit lefties at an .870 OPS across 356 at-bats. Alignment is not an edge, no line had posted when the models locked, but unanimity this tight is unusual.

Tonight's oddity is Colorado at the Dodgers, where the Rockies' season splits into a paradox. Games where Colorado faces a lefty average 10.7 total runs, and LA starts lefty Eric Lauer. But Colorado wins just 22% of road games against left-handed starters (54-game sample). The bats travel, the wins do not, and the base model calling this near a coin flip (50.6% LA) reads generous to Colorado against those splits. Coors inflates that runs number, so carry the caveat.

And the slate has bookends: the model projects 12.3 total runs for Houston at Washington (the Nationals average 10.2 at home against righties, 147 games) and just 6.3 for Toronto at San Francisco, a 9:45 start on a mild bay night at a park suppressing homers at a 0.76 factor. Same sport, same night, a six-run gap in expected environment. If the St. Louis heat shows up, Jackson Chourio (1.079 OPS at Busch in 60 at-bats, .967 in hot weather across 306) is the bat to watch in Milwaukee's spot, though that one stays conditional until a hot first pitch actually posts.

One more thing

The board changes every night, and this newsletter is how you keep up without living on the site. One email a day, the calls before they resolve and the grades after, free, no spam. Sign up here.