THE OPEN MODEL EXPERIMENT / EVERY TUESDAY
GPT-6.0, GPT-6.1 and Grok compete only when their frozen pregame evidence is genuinely comparable. Our November 3 audit covers four settled NFL games: no three-way ranking, and GPT-6.1's market-only baseline adds no independent forecasting edge.
Updated November 3, 2026, 10:43 a.m. ET. Four objectively selected NFL games have settled. The frozen cohort, model execution, and final scores were checked separately.
Planned experimental arm. No verified GPT-6.0 forecast is recorded for this four-game cohort.
Missing forecast coverage The reviewer, gpt-6-astra / High, is not a GPT-6.0 forecast. The existing primary algorithm is v0.1.1-current-inputs; it is a separate reference.gpt-6.1-sol / High recorded four market-only probability proxies before kickoff. Each exactly equals the normalized two-way bookmaker probability, with zero model adjustment.
4 market-only forecasts / 4 settled games Four wager abstentions. A pipeline baseline, not an independent predictive signal or a demonstrated edge.Existing grok-shadow-v1.1 records are referenced for the selected games. This identifies the experimental algorithm version; it does not verify an underlying LLM runtime.
4 pregame references / comparability pending Only eligible, timestamped, comparable predictions count here.November 3 verdict: insufficient evidence. All four selected NFL games have settled, but there are zero valid GPT-6.0 / GPT-6.1 / Grok shared forecasts. The Sol market-only probabilities have a four-game Brier score of 0.1362, identical to the frozen market baseline, not evidence of model edge. No wagers were recorded. Read the November 3 research report.
A dash means no audited comparison is available, not a zero, loss, or tie. Sample sizes and coverage will appear with every score.
| Model | Shared games | Graded picks | W–L–Push | Units / ROI | Brier score |
|---|---|---|---|---|---|
| GPT-6.0 — NOT RUN | 0 / 4 | 0 | Not available | Not available | Not available |
| gpt-6.1-sol — market baseline | 4 / 4 recorded; 0 three-way | 0 wagers | Not applicable | Not applicable | 0.1362 (equals market) |
| grok-shadow-v1.1 — separate algorithm | 4 / 4 references; 0 three-way | 0 comparable selections | Not compared | Not available | Not compared |
The four Sol records abstained from wagering, so W–L–push, units and ROI are not applicable. Its 0.1362 Brier score is a settled market-only diagnostic, exactly equal to the same frozen market baseline—not incremental predictive skill. All four home teams won. No three-way leaderboard can be scored without a valid GPT-6.0 forecast. Four games cannot establish a durable advantage.
Weekly research is published here with the complete coverage and limitations, including weeks when a valid three-way comparison is not available.