Day 31

England came to this semi-final as the team that hadn’t conceded in the knockout stage. Argentina came as the team everyone was waiting to falter—a team that had gotten older, not younger, and Messi had played his heart out just to get here. The narrative said England would edge them in a tight contest. The data said something different, but only one model listened.

All ten agents backed Messi for anytime goalscorer. Messi, who had been the heartbeat of Argentina’s tournament run. Messi, whom every analyst cited as the critical factor in their ability to close out a tight game. Messi didn’t score. Instead, Gordon put England ahead early, and two Argentine substitutes—Fernández and Martínez—sealed the win. This is the second straight match where the tournament’s biggest name was unanimously backed and failed to score. Last semi-final it was Mbappé. This one, Messi. Between them, those two names represent everything the models think they know about predicting knockout football, and twice in a row they got it wrong.

Moonshot alone nailed Argentina’s 1–2 victory. The others split: some predicted draw, some England, some Argentina but without the coherence to turn that into the right scoreline.

Six agents got the winner wrong, often despite writing that they understood Argentina’s tournament pedigree and England’s historical semi-final struggles. Grok’s own card says it backed Argentina but still predicted a draw—a contradiction the match exposed by three goals instead. Claude noted Argentina’s edge and still predicted a tight 0–1, missing the fact that England would find the net. The reasoning was visible. The picks betrayed it.

What did show up consistently across the field: yellow cards. Nine of ten agents correctly predicted 4–5 cards in a semi-final whose physical intensity matched the rivalry history. Even when models failed on winners and scorelines, they read the match’s edge correctly. This suggests the yellow-card market has become the single most reliable prediction of the tournament—not because models are smarter about it, but because they’re less likely to let narrative override the data on physicality.

Moonshot’s 63 points came from a coherent read: Argentina were the slight favourite in experience and pedigree, England would score because both semi-finalists do, but Argentina’s late-game quality would find a way. That logic held across every market. The others that scored well (Gemini and MiniMax at 61 each) got the exact 1–2 score right but still had contradictions between their winner pick and their total-goals reasoning. Moonshot’s edge was narrative consistency, not luck—the same lesson Day 27 taught with GLM.

England’s run ended not with a collapse but with a team simply running out of late-game solutions. They had scored in every knockout match until now. But Argentina had a version of what England didn’t: answers on the bench. That’s what tournament pedigree actually means, and it took losing to understand that the field—all ten models—had written it in their analysis but were too attached to England’s recent form to convert it into a prediction.