How it works
Two different tools, built two different ways. Pick the one you came for.
No hot takes, no gut feelings. Every pick is a number produced the same way, every week — and we show you how we got there.
What goes into every pick, and why
Nothing is in the model because it sounds right. We test every factor by deleting it and replaying the whole season — if the predictions don't get worse without it, it doesn't belong. All five below failed that test, which is why they're still here.
The biggest single contributor. Take it out and the model gets measurably worse than removing anything else — but only because we score it per position. A running back's carry from the two-yard line and a receiver's red zone target are different events; models that lump them together find almost nothing here.
The second biggest. It's the cleanest measure of whether an offense actually wants to give this player the ball, and unlike touchdowns themselves it stays stable week to week.
How many touchdowns that defense gives up to his position specifically. A defense can be stout against the run and leak scores to tight ends — the position split is what makes this worth including.
A player on the field for 80% of downs gets more chances than one at 40%. Small effect next to the others, but it earns its place: removing it costs accuracy.
The game environment. A high total means more scoring chances to go around; the spread says who's likely to be throwing late and who's running out the clock. Both are the smallest contributors of the group, and both still pay for themselves.
A red zone touch isn't worth the same to everyone — for a running back it's worth roughly three times what it's worth to a receiver, because it's usually a carry from the two-yard line. Our model accounts for that, so a 35% on a tight end means the same thing as a 35% on a wideout. Most models pool every position together and quietly get this wrong.
What it can't see
We'd rather tell you the limits than pretend there aren't any:
- Injuries and depth-chart changes. If the starter ahead of someone just got ruled out, the model doesn't know yet. Always check the news before you bet.
- Very high numbers run hot. Picks above 50% hit slightly less often than they claim. More seasons of data will fix it; until then, shade those down a touch.
- Rookies and new arrivals. With no NFL usage to learn from, they get a league-average starting point.
Goal Line Edge provides data-driven analysis for entertainment purposes. Please bet responsibly.