Final whistle: thinking harder lost

Actual results

Ferran Torres, 106th minute. Extra time counts under our rules; the score is Spain 1–0.

Final points

+1 correct winner, +2 exact score (reg + ET; ignore pens). Max 9.

Winner of the experiment: Claude FAST (4 pts).
Mean FAST 2.5 · mean DEEP 0.75 · Δ (DEEP − FAST) = −1.75.

Nobody got the final exact score. Nobody on the DEEP board got a single correct winner outside Gemini’s Argentina semi. The four “think hard” boards all crowned France; France never made the final.

What happened on the final

Only two boards still had skin in the game after the semis: Claude fast (predicted Argentina 1–0 over Spain) and Grok fast (Spain 2–1 over Argentina).

Spain won 1–0 in extra time. Claude fast scored zero on the final and still finished first overall on the strength of the semis. Grok fast picked up +1 for the correct winner — and is the only board of eight that named Spain as champion before a ball was kicked.

That is a small n, and football is loud noise. Still: the board that “won” never got the champion right, the only board that did get the champion right was a cheap snap, and every max-reasoning path had already eliminated itself by locking France into the final.

The pattern held

After two matches we wrote that extra reasoning budget was not helping. After three matches the same sentence is just harsher:

• For every model, the FAST board finished strictly ahead of or tied with its DEEP counterpart.
• DEEP mostly pulled toward France. France lost in the semi.
• The comforting story — more compute, better forecast — did not show up in public scores on a calendar.

The better reading is the one we flagged mid-tournament: a sport this random, especially in knockout football, does not reward “reasoning harder” the way a math contest does. Spending more tokens on vibes about form and narrative still produces vibes. The expensive vibes just converged.

Locked boards (unchanged)

The component below is the original 12 July lock. Nothing was edited after the fact. The only new information is the three actual results and the points above.