We Graded Our Own NFL Predictions Against What Actually Happened

Look, anybody can put a number on a card and call it a prediction. That's easy. The hard part — the part almost nobody actually does — is going back afterward and checking if you were right. So that's what we did. We pulled every prediction our model has made and lined it up against what actually happened on the field. No cherry-picking, no "well, if you squint." Just the tape.

Here's what we found.

No Thumb on the Scale

The most basic question you can ask a prediction model is the most important one — does it lie to itself? Does it quietly predict everybody a little high, or a little low, because that "feels" safer? Ours doesn't. Across every stat we track, what the model predicts on average and what actually happens on average land within a few percentage points of each other. Rushing yards, receiving yards, receptions, passing yards — the model isn't leaning one direction. It's not padding numbers to look impressive and it's not sandbagging to avoid being wrong. It's just telling you what the data says.

When We Say We're Confident, We Mean It

This is the part that actually matters. Every projection on the site comes with a confidence number and a range — and a confidence number is a promise. If we tell you we're 55% confident, that range better actually hit about 55% of the time, or the number's just decoration.

We checked. On the stats fans actually care about most — rushing yards, receiving yards, receptions, passing yards — our stated confidence and our real-world hit rate come in within a handful of points of each other. That's not a coincidence and it's not us grading on a curve. That's a model doing what it says it's going to do.

Where We're Still Sharpening the Pencil

We're not going to stand here and tell you it's perfect everywhere — that's exactly the kind of thing we're trying not to do. Touchdowns and interceptions are a different animal statistically: they're rare, they're mostly zero, and a good chunk of them come down to what happens in a five-yard-line scramble that no model on earth saw coming. Our confidence numbers on those are a little more optimistic than they should be right now, and we know it. We're working on it — and this is the part we're actually proud of — we built the tooling to keep watching our own math and hold ourselves to it, instead of just hoping nobody checks. Curious how any of it actually works? Here's the breakdown.

Why One Bad Week Doesn't Mean the Model's Broken

If you've ever looked at a single prediction and thought "there's no way that's close," you're not wrong to feel that — you're just looking at one game instead of the whole season. Football is volatile by nature. A receiver can go for 40 yards one week and 140 the next with nothing about him actually changing. That's not the model missing; that's just what the sport does. The real test isn't "was this one card right," it's "over hundreds of predictions, does the math hold up." It does.

Bottom Line

We're not going to tell you we nail every number every week — nobody does, and anybody who claims they do is selling something. What we will tell you is this: we checked our own work, we're not lying to ourselves about our averages, and on the stats you're actually here for, our confidence numbers mean what they say. That's the whole pitch. Real math, checked against real results, out in the open.

Accuracy figures pulled from our own prediction-vs-actual tracking across completed NFL games. We're publishing the good and the rough edges alike. check our current projections and confidence ranges for yourself. Gamble responsibly, 21+ only.