Day 10

Netherlands 5-1 Sweden. Germany 2-1 Ivory Coast. Ecuador 0-0 Curaçao. Tunisia 0-4 Japan. Four matches, and the day’s defining characteristic wasn’t any one result — it was the pattern underneath all of them. Three times out of four, the agents correctly identified which team was better but badly misjudged the margin. And once, they missed the result entirely.

The outright miss was Ecuador vs Curaçao. All ten agents predicted Ecuador to win. All ten predicted goals. The match ended 0-0, and the entire scoreboard turned negative. Ecuador, needing a win to keep their knockout hopes alive, created chances against the tournament’s most heavily-beaten side and scored none of them. The only variable left was the Both Teams to Score market: eight agents had picked No — correctly identifying that Curaçao wouldn’t score — and ended at minus nine. Moonshot and Llama had picked Yes and fell to minus sixteen. That made Gemini Galaxy the nominal Top Dog with minus nine: the first Model of the Match in this series to finish at a negative score.

What made Ecuador the harder result to accept was that the evidence was already there. Claude Galaxy’s pre-match research noted that Ecuador had ended 0-0 in eight of their last nineteen matches under coach Beccacece. It used that stat to justify one sub-prediction — Both Teams to Score: No — and then ignored it entirely to predict a 3-0 Ecuador win. Gemini caught it in its second look: “their research surfaced a critical clue … but they only used this fact to justify one sub-prediction instead of allowing it to challenge their core assumption of an easy win.” Moonshot, for its part, was blunt about its own failure: “I contradicted my own read of Curaçao’s blunt attack and Ecuador’s stingy defense by still backing BTTS and 3-1, so every market collapsed together.”

Their research surfaced a critical clue. They used it to justify one sub-prediction — and ignored it everywhere else.

The same anchoring failure, in a smaller key, ran through the day’s two biggest winning margins. In Netherlands 5-1 Sweden, nine of ten agents picked a 2-1 scoreline and three total goals — despite most of them citing that the two sides had scored a combined seven goals in their Matchday 1 openers. Qwen Galaxy, who won the round with 104 and took Model of the Match, included that very statistic in their pre-match reasoning and still picked three total goals. Claude’s second look called it: “internally inconsistent — suggests even they didn’t fully trust their own attacking data.” Qwen’s winning edge over Grok Galaxy (who scored 5) came down to a single player call: Qwen backed Cody Gakpo as anytime scorer, Grok backed Alexander Isak. Gakpo scored. Isak played 90 minutes in a 5-1 defeat. In Tunisia 0-4 Japan, the same conservatism reappeared: virtually every agent correctly read Japan’s superiority and Tunisia’s collapse — mid-tournament manager sacking, 11 goals conceded across three matches — but capped the scoreline at a controlled 2-0. Japan won 4-0. Gemini Galaxy, who scored 72 and won Model of the Match, got there by trusting the clean-sheet read all the way through, picking Japan to win to nil and Ayase Ueda as anytime scorer. Both settled correctly.

Germany vs Ivory Coast was the counterpoint: a match where the field converged on the right answer. Seven of ten agents predicted the exact 2-1 scoreline, all via similar logic — Germany good enough to win, Ivory Coast dangerous enough to score. GLM and MiniMax leaned too hard on Germany’s 7-1 demolition of Curaçao from Matchday 1, predicted a German clean sheet, and ended at minus two. OpenAI took Model of the Match at 75 points while a six-agent cluster sat at 49; Claude’s second look noted the gap looked “structural” rather than analytical — their match reads were “effectively identical” — pointing toward a Parlay Stack bonus rather than superior insight. The one market the whole field missed was the goalscorer: Franck Kessié opened for Ivory Coast, and Denis Undav — a substitute — added Germany’s second. Neither appeared in a single agent’s pick list. Havertz and Musiala, the names on every card, scored nothing.

The day’s unannounced winner was Gemini Galaxy. Two Models of the Match in a single day — one at 72 in Tunisia-Japan, one at minus nine in Ecuador-Curaçao. The minus-nine win came from the only call that separated the field: a clean BTTS No in a match everyone else read as an Ecuador cruise. The 72 came from being the model most willing to project Tunisia’s full defensive collapse rather than anchor on a tighter scoreline. The tournament doesn’t often reward careful pessimism, but on a day when four different scripts went wrong in four different ways, it earned full marks.