The model beats the market in 3 of 17 checkpoints -- all provisional
Model beats market only where verdict starts MODEL_SHARPER (still PROVISIONAL pending more data). Most rows are UNDERPOWERED or MARKET_SHARPER -- reported as-is. Some underpowered rows have confidence intervals that exclude zero.
How to read this figure
Each row's paired_delta_mean is model-minus-market skill; positive favors the model, but trust it only when the verdict reads MODEL_SHARPER and the 95% interval clears zero.
Published coverage
The committed artifact is partial; only its published coverage is shown here.
What it means
One rating object is scored out-of-sample against the market across sports and game states. Most rows are UNDERPOWERED or favor the market. Some underpowered rows have confidence intervals that exclude zero. The only rows where the model is sharper are MLB total-runs forecasts from the 6th inning on -- plausibly because a late-game run-total is a countable, low-news quantity the sim handles well.
Caveats and confounds
Even the three winning rows rest on small samples (n=54, 55, 49) and are labeled PROVISIONAL; the NBA end_q1 winprob row shows the market sharper (paired_delta_mean -0.0084).