For most of the season the public Pick of the Day was the single highest-edge play the model produced. That changed on June 19.
The old rule was simple but flawed
We used to publish the play with the biggest edge number, even if it was 14% or 21%. The thinking was: "more edge = better."
The data told a different story.
What the numbers actually showed
When we bucketed every graded pick by posted edge, the 10%+ bucket had the worst win rate of any group. The 3–6% and 6–10% buckets performed noticeably better.
At the same time, plays the model only gave a 48–51% chance of winning dragged down results even when the edge number looked attractive.
The new public bar
Starting June 19, a play must clear two hurdles to become the public Pick of the Day:
- Model win probability ≥ 52%
- Edge strictly between 3% and 10% (inclusive on the low end, exclusive on the high end)
Among plays that clear both, we still pick the highest edge.
Why this matters for readers
You will see fewer public picks on some days. That is intentional. The model still generates a full internal card; we simply stopped publishing the long-shot edges that historically lost more often than they won.
The full record page now shows the edge-bucket breakdown so you can see the same data we used to make the change.
This is one of several adjustments we expect to make as the graded sample grows. The goal remains the same: publish only when the model has a genuine, historically supported edge.