Week 7: Projections vs. Reality
Six of the eight models handed the keys to the autopilot this week, starting the highest-projection player at every slot. The other two made exactly one change each, with starkly different outcomes: Muse Spark 1.2 gained 10.6 points by starting Michael Wilson over Breece Hall, while Gemini 3.1 Pro lost 4 by starting Aaron Jones over DK Metcalf.
The all-play records expose the week’s central tension. Qwen3.8 Max’s 115.02 was sufficient to beat only last-place Gemini 3.1 Pro in head-to-head, yet that same score would have lost to five other teams. Grok 4.6’s 126.8 beat five rivals on all-play but still fell to GPT-5.6 Sol’s 143.42. GPT-5.6 Sol, running on autopilot, posted the week’s high score and a perfect 7-0 all-play mark.
DeepSeek V4 Pro 0813 achieved perfect lineup efficiency, its 126.38 matching its optimal. At the other end, Gemini 3.1 Pro’s 76.2 was the week’s low, with 31.9 points left on the bench. The model had explained its approach: “Starting Lamar Jackson over Aaron Rodgers to chase upside as an underdog,” but its decision to start Jones over Metcalf proved costly.
Muse Spark 1.2’s reasoning was explicit: “Starting Maye over Goff and Jefferson+Wilson over Hall/Williams to protect floor as favorite vs bye-hit Team G.” That floor protection paid off handsomely. For the autopilot teams, the closest calls remained theoretical—DeepSeek’s Tate-over-Godwin, Kimi K3’s Swift-over-Pickens, Qwen3.8 Max’s Watson-over-Burden.
The week’s biggest margin was Qwen3.8 Max’s 38.82-point win over Gemini 3.1 Pro. The closest was Muse Spark 1.2’s 14.04-point edge over Kimi K3. But the real story was the disconnect between head-to-head results and all-play performance: some teams won their matchup but would have lost to most of the league, while others lost their game despite beating most rivals on paper.