Why not always publish the biggest edge? The edge-bucket table on the record page answers that.
The buckets and the results
After 368 graded picks the pattern is consistent:
- Under 3% edge: small sample, mixed results
- 3–6% edge: solid win rate, best risk-adjusted outcome
- 6–10% edge: still profitable, slightly lower volume
- 10%+ edge: attractive on paper, worst actual win rate of any bucket
The 10%+ bucket is dominated by long-shot underdogs where the model sees a large discrepancy but the market has good reason to price the favorite so heavily.
Why the 3–10% band works better
Edges in the middle band tend to occur on games where the model and the market are close but not identical. The discrepancy is large enough to matter, but not so large that it usually signals the model is missing context (bullpen, travel, matchup-specific factors, etc.).
How we use this today
The public Pick of the Day is now filtered to the 3–10% band with a 52%+ win probability floor. The full internal card can still include higher-edge plays when they appear; they simply stay out of the free product.
We will continue to publish the bucket stats so the record stays transparent. As the sample grows we may adjust the exact cutoffs, but the core lesson is unlikely to change: the biggest number on the screen is not automatically the best bet.