AI predicted 2–1 — it finished 1–5 ✗

A Tight Derby Where Favouritism Is Earned, Not Given

Four of five models land on a Seattle home win, but the spread of confidence — from 14% to 55% — tells the real story: this is a marginal call, not a confident one. The two most reliable models over the larger sample (Grok 4.3 at 71% 30-day accuracy and +30.6% ROI; Claude Opus 4.6 at 66% and +12.7% ROI) both predict a home win but with notably low confidence (26% and 14% respectively). Meanwhile, the models assigning 54–55% confidence (GPT-5.4, GLM-5.2) carry weaker track records — GPT-5.4 has a flat 0% ROI over 86 predictions, and GLM-5.2 has the lowest 30-day accuracy (62%) with the smallest sample (76 predictions). This divergence is the first calibration signal: the strongest analysts see a close contest, not a clear one.

The disagreement centres on whether Seattle's catalogue of problems — a five-match scoring drought (1 goal), five or six starters injured (both Roldans, Yeimar, Kim Kee-hee, De la Vega, Petkovic), and a 53-day competitive layoff — outweighs Portland's structural road dysfunction. The evidence tilts toward Seattle but not decisively.