Opening night, September 5, 2012. Cowboys at the defending-champion Giants. Kevin Ogletree, Dallas’s third receiver, catches eight balls on eleven targets for 114 yards and two touchdowns. Twenty-seven fantasy points, out of nowhere, in the first game of the year.
I put in the waiver claim. I’m pretty sure I dropped somebody to do it. And over the other thirteen games of that season Ogletree caught 24 passes for 322 yards and two scores, about three targets a game, with two games where he didn’t catch a ball at all (all of it from nflverse, 2012 season). I had found my breakout gem, and the gem was one game long.
I’ve done this more than once. It always happens early in the season, when the sample is tiny and every big line feels like a discovery. So this is the first piece in a series about how I read the numbers now, and, more to the point, how the model on this site reads them. I’ve built a thing, with a lot of help from AI, and I’m testing it in public every week against the betting market and the box score. The first step in that is being honest about which stats deserve trust.
Because not every fantasy football statistic is equally useful, and even a useful one can send you the wrong way when you read it on its own. I sort stats into three buckets: the ones I trust, the ones I use with context, and the ones I don’t trust alone. The sorting isn’t a matter of taste. There’s a test behind it.
The test: does it stick around?
A stat earns trust if it predicts itself. If a player’s number in September tells you roughly what his number will be in November, you can build on it. If it doesn’t, you’re building on sand, however good it looked at the time.
So I ran that test on the 2025 season. Take every running back, receiver and tight end who played at least six games in each half of the year (202 of them), average each stat over the first nine weeks, average it again over the back nine, and see how well the first half lines up with the second. A correlation of 1.0 means the first half told you everything. Zero means it told you nothing. Here’s the ladder, from nflverse data, half-PPR scoring.
| Stat, per game | Predicts itself |
|---|---|
| Carries | 0.94 |
| Air yards | 0.87 |
| Target share | 0.86 |
| Targets | 0.82 |
| Receptions | 0.79 |
| Red-zone opportunities | 0.77 |
| Fantasy points | 0.76 |
| Touchdowns | 0.53 |
| Catch rate | 0.25 |
| Yards per target | 0.17 |
| Yards per carry | 0.00 |
Read that top to bottom and you’ve basically read my whole philosophy. The stuff at the top is opportunity. The stuff at the bottom is efficiency. Fantasy points sit in the middle, which is exactly where they belong: a real signal with a lot of noise mixed in. And yards per carry, the number every broadcast puts on screen, carried literally nothing from one half of the season to the other.
That’s the framework in one sentence. Opportunity leads to usage, usage leads to production, production leads to points. Every step down that chain adds luck, so the further upstream you read, the steadier the picture. I lean toward consistency over boom-or-bust anyway, and this is why: the consistent stuff is the stuff at the top of that table.
The stats I trust
Snap share. Being on the field is opportunity, full stop. A player can’t be targeted from the bench, and a coach who keeps a guy out there for 80% of the snaps is telling you something he won’t say in a press conference. Snap share is the first thing I look at for anyone I’m thinking about picking up.
Targets and target share. A target is a coach deciding, on purpose, to get a player the ball, and nothing in fantasy is more repeatable than a coach’s intentions. Target share matters more than raw targets because it survives a slow offensive day: seven targets in a game where your team only threw 22 times is a bigger vote of confidence than nine out of 45. Jaxon Smith-Njigba took 46% of Seattle’s targets in week one. Amon-Ra St. Brown took 36% of Detroit’s on 14 looks (nflverse, 2026 week 1). Those numbers don’t need a touchdown attached to mean something.
Carries. The single steadiest thing in the table. Backfields are decided in meeting rooms on Wednesday, not on the field on Sunday, and coaches don’t change their minds easily. A back who had 15 carries a week in the first half of 2025 almost always had about 15 in the second half.
Red-zone opportunities. Touchdowns are noisy. The chances to score them are much less so. Red-zone targets and carries held up at 0.77 across 2025 while actual touchdowns came in at 0.53. If a player keeps getting the ball inside the 20 and keeps not scoring, he’s usually about to. If he’s scoring without those chances, enjoy it, and keep an eye on the sell window.
Air yards, for receivers. How far the ball travels toward a player measures what kind of role he’s got. A receiver with 130 air yards a game is being asked to make big plays whether or not they connected, and that role tends to stick.
Role, especially for rookies. This one isn’t a stat so much as the thing the stats are describing. With a rookie receiver it’s all about opportunity. Is he the WR2 in a good offence, or the WR3 or WR4? The 2 is going to get his chances. The 3 or 4 needs an injury, or has to earn it a snap at a time, and until one of those happens the box score is going to be quiet. That was Ogletree’s real problem in 2012. He wasn’t bad. He was the third guy behind Dez Bryant and Miles Austin, and one big night didn’t change that.
This is also the part I got right once. I was a Cooper Kupp fan before the breakout, because the opportunity was already there and the news around Stafford coming in made it better. In 2020 Kupp saw 124 targets and caught 92 balls for 974 yards, with three touchdowns. In 2021 it was 191 targets, 145 catches, 1,947 yards and 16 touchdowns (nflverse). The targets were there a year early. The touchdowns just hadn’t shown up yet. That’s the whole idea in one player.
The stats I use with context
Fantasy points per game. This one surprises people. Points per game is a decent stat, and at 0.76 it’s plenty stable. But it’s a summary, and a summary hides the ingredients. Two receivers can both average 13 points a game, one on nine targets a week and the other on six targets plus a hot streak in the end zone, and they are not the same player going forward. Points tell you the score. They don’t tell you how the score was built, and how it was built is what carries over.
Snap share without targets. I said I trust snap share, and I do, but it’s the one stat in the top bucket that needs a second look. Michael Pittman played 88% of Pittsburgh’s snaps in week one and drew three targets. Rashod Bateman played 78% of Baltimore’s and drew one. Chicago’s Colston Loveland was out there for 87% of the snaps and saw two balls (all nflverse, 2026 week 1). Snap share tells you a player is on the field. Target share tells you whether anyone is looking for him. You need both, and when they disagree, the targets usually win.
The stats I don’t trust alone
Don’t misunderstand me here. Touchdowns and yards per carry don’t lie, and it’s not that they don’t matter. It’s that you have to see the whole picture, and these are the numbers that most tempt you not to.
Say a running back gets 20 carries for 60 yards and no touchdowns. Looks bad. Three yards a carry and a goose egg. But you might not be seeing the whole picture from the box score. Maybe the defence was elite. Maybe they stacked the box all day. Maybe he had two carries from the one and got stuffed. What I see is 20 carries. That’s opportunity, and opportunity is the part that comes back next week.
Touchdowns. The biggest number on the stat sheet and the flakiest. Touchdowns held at 0.53 from one half of 2025 to the other, which is roughly a coin with a slight lean. Among the top 24 receivers by points last season, an average of 22% of their fantasy points came from touchdowns, but the spread ran from 11% to 44% (nflverse, 2025). Davante Adams was at the top of that range, 44% of his production from scores, on about eight targets a game. CeeDee Lamb was at the bottom, 11%, on nine targets a game. Adams needed the touchdowns to keep his number. Lamb didn’t. That’s why I look at where the points came from and not just the total.
Week one gave us the live version. Dallas Goedert scored 21.7 on five targets and two touchdowns. Dalton Kincaid scored 15.5 on six targets and no touchdowns, with 130 yards (nflverse). If I had to bet on which of those lines repeats, it’s Kincaid’s.
Yards per carry. Zero on the table. A back’s efficiency in the first half of 2025 told you nothing about his efficiency in the second half, because it moves on one long run, which is why the weekly leaders are usually somebody with four carries. Yards per carry is a fine description of what happened. It’s just not a description of the player. Same with catch rate, same with any efficiency number: they describe the day, and the day is over.
What the model does with all this, and what’s coming
The model on this site reads the box score in the order that table suggests. Opportunity first: is the player on the field, and what role is he playing? Then usage: how much of the offence actually runs through him? Then it works forward to production and, last of all, points, with efficiency weighted lightly on purpose, plus the game situation and a check on whether what it’s seeing can last.
I’m not going to tell you it’s right. That’s what the scorecard is for, and it’s why this is a series and not one article. Over the next few weeks I’ll walk through one layer at a time: how the model decides who has the opportunity, and the role-change idea I tested and had to leave out; what it does with target share and carries; why it barely trusts efficiency, with the backtest numbers that made that call; what a Vegas line actually adds to a projection, and what it costs; and, at the end, the scorecard itself, with every week’s grade next to what the betting market and the plain box score would have done instead. If it’s beating them you’ll see it, and if it’s getting beaten you’ll see that too.
Which brings me back to Ogletree. Eleven targets and two scores on opening night looked like a role. It was a night. The version of me that reads the top of the table before the bottom waits a week, looks at who’s ahead of him on the depth chart, and probably keeps the player I dropped.
The best fantasy managers aren’t ignoring the box score. They’re learning to look past it.