Sierra Leone 2–3 Zimbabwe: AI predicted 1–0 ✗

A tight, low-scoring opener decided by context, not class

Claude Opus 5, Grok 4.6 and GPT-5.6 all anchor on this, and it is genuinely decisive. It strips out the crowd, the poor pitch, the humidity and the travel fatigue that usually make CAF qualifiers so home-skewed — Sierra Leone keep the administrative 'H' and little else.

Where the panel disagrees — and how it resolves

The one genuine split is GPT-5.6's 0-1 Zimbabwe call, built on the Warriors' superior central-midfield core (Nakamba, Munetsi, Rinomhota) and Ramón Catalá's transition threat. It's a real argument, but it runs into two problems the other four models identify: Catalá only took charge in late August with players still arriving and Zemura injured, whereas Gomes Da Rosa has been in post since June and held his squad in Morocco preparing on site; and Zimbabwe have a home fixture against DR Congo four days later, which strongly incentivises a conservative, point-taking setup on matchday one. Zimbabwe's individual quality is real, but their operational chaos plus strategic contentment outweighs it here. The panel majority's home lean survives this challenge, though the fact that the strongest dissent comes from squad-quality analysis keeps confidence pinned at 0.38 — genuinely low.

The second disagreement is subtler: whether the neutral venue cancels Sierra Leone's edge (Claude, GPT) or merely dilutes it (Kimi, GLM, Grok). The resolution lies in Sierra Leone's June double-header with Liberia — 1-0 at home, 1-3 away against the same opponent four days apart — a four-goal swing driven almost entirely by venue. Even a Morocco-based 'home' game keeps them on the comfortable side of that split, while their 107-day layoff and blunt attack (one goal per friendly against modest opposition) cap the margin. That is why the panel lands on 1-0 rather than anything more ambitious: the Leone Stars are the likelier winners of a game neither side can realistically win by more than one.