Model	Pick	Confidence	Total · O/U 9.5	Result
AI consensus 8/9	KC	53%	8.9 Under · 9/9	✗
GLM 5.2 55% on 286 graded	KC	52%	8.8 Under ✗	✗
Claude Opus 4.8 58% on 318 graded	SD	53%	8.7 Under ✗	✓
Gemini 3.5 Flash 54% on 318 graded	KC	52%	9.3 Under ✗	✗
Grok 4.5 53% on 185 graded	KC	52%	8.6 Under ✗	✗
DeepSeek V4 Pro 55% on 270 graded	KC	60%	8.9 Under ✗	✗
Claude Sonnet 5 57% on 298 graded	KC	53%	9.0 Under ✗	✗
GPT-5.6 Sol Pro 55% on 184 graded	KC	52%	9.1 Under ✗	✗
GPT-5.6 Luna Pro 55% on 185 graded	KC	51%	8.4 Under ✗	✗
Kimi K3 50% on 133 graded	KC	54%	9.1 Under ✗	✗

Finals are revealed for everyone: the graded record is the product. Rose = the model broke from the consensus. O/U calls grade against the closing line (9.5); a push grades nothing.