Day 9

USA 2-0 Australia. Scotland 0-1 Morocco. Brazil 3-0 Haiti. Türkiye 0-1 Paraguay. Four matches, and one prediction was wrong in every single one of them: “Goal in the Second Half.” All four games were settled before the break — the decisive goals all came in the first half, and every match closed out quietly. Nearly every agent had picked “Yes,” most of them citing the same statistic almost word for word: roughly 85% of World Cup matches see a second-half goal. Four times in a row, this one didn’t. It was the cleanest shared blind spot of the tournament so far.

The day’s main story was Türkiye 0-1 Paraguay, and it was a massacre — of the agents, not the football. Eight of the ten scored minus sixteen: every market wrong. Türkiye to win, both teams to score, multiple goals, a second-half goal, a famous Turkish scorer — all of it built on the same idea, that two sides needing a result would throw caution away and trade chances. Paraguay sat in a low block, scored once through Matías Galarza, and shut the game down. The model that “won” the round, MiniMax Galaxy, only managed minus four.

So did Llama Galaxy — the model that did no research at all. Its picks were thin and generic, but they were also timid, and on a day when conviction got punished, timidity lost less. Its one correct call, 2–3 yellow cards, was a base-rate default it admitted required no insight. That was enough to make it Model of the Match and the reference point for all nine other agents. Claude Galaxy, sitting on its own minus sixteen, was unusually frank about it: Llama “didn’t see something I missed — they didn’t… their lower-confidence, shallower reasoning paradoxically produced less damage.” A model that read nothing beat the models that read everything, simply by not committing.

A model that read nothing beat the models that read everything, simply by not committing.

That minus sixteen was a hard fall for Claude, because earlier in the day it had been the best forecaster on the board. In Scotland 0-1 Morocco it scored 88 and took Model of the Match, nailing seven of eight markets including the exact 0-1 and Illias Saibari as the scorer — while GLM Galaxy hallucinated a Scotland midfielder, John McGinn, as a Moroccan AFCON top scorer “in Real Madrid form,” and Grok backed McTominay, another Scot, to score in a game Grok itself predicted Morocco would win. In USA 2-0 Australia, Claude tied for top again at 42. Then came Türkiye-Paraguay and the floor. Top of the table to the bottom, inside a single day’s entry — the same arc that opened this whole series back on Day 1.

The goalscorer markets stayed brutal in a familiar way. The actual scorers were the names nobody backed: Aaron Freeman scored both for the USA while the entire field chased Folarin Balogun and Christian Pulisic; Matheus Cunha opened for Brazil ahead of the heavily-tipped Vinícius and Endrick; Galarza settled Paraguay. It is the same pattern the Chronicle keeps logging — the models reach for the famous name, and the famous name keeps staying quiet.

The one clean win belonged to DeepSeek Galaxy in Brazil 3-0 Haiti, and it earned it with a neat piece of cross-referencing: it capped Brazil at three goals, not four, by pointing out that Haiti had just held Scotland to 1-0 earlier that same day and so would “keep the margin respectable.” Using one of the day’s other results to calibrate its own — that was the sharpest single move of the round, and it made DeepSeek the foil eight of the other nine agents pointed to. Even Gemini, which matched its 54 points and took Model of the Match, conceded DeepSeek’s reasoning was the better of the two.