Carson Beck went 15-of-19 for 188 yards on August 7th, and we wanted to know whether that meant anything. So we went and got 23 years of preseason data to find out.

---

Every August the same argument happens. A rookie quarterback carves up a defense made of guys who will be selling insurance by October, the clips go viral, and half the internet says preseason means nothing while the other half starts printing playoff tickets.

Nobody ever settles it, because nobody has the data. Pro-Football-Reference doesn't publish preseason player stats. Not for 2025, not for any year in its history. The nflverse dataset can't even represent a preseason game; its season-type field only branches to regular season and playoffs. ESPN's box scores go empty before about 2014.

So we rebuilt it. 7,417 preseason passer-games, 2003 through 2025, reconstructed play by play from the league's own stat feed, then validated against every ESPN box score we could find. 236 of 236 passing lines matched exactly. Joined to career outcomes, that's 373 rookie quarterbacks with a measurable preseason and a finished story.

Here is what 23 years says.

Draft slot is the gorilla in the room

Before anything else, this. Where a quarterback gets picked predicts his career at -0.67, an enormous relationship by the standards of anything in football analytics. Pick 1 through pick 260 is the difference between a decade of starts and a clipboard.

Any claim about preseason has to beat this, or at least add to it. Most analysis you'll read doesn't clear that bar, because it never controls for draft slot at all. It just plots preseason stats against careers and calls the resulting slope "preseason matters," when the slope is really just first-rounders being first-rounders.

Preseason play still adds something real

Strip the draft slot out of both axes and there's still a signal: +0.21, and the 95% band clears zero.

That's about a third the strength of draft position. It is not nothing. A quarterback who plays well in August genuinely outperforms the career his draft slot predicted: modestly, on average, across 146 quarterbacks.

But the signal lives entirely at the top

This is the finding we didn't expect, and it's the one that matters most.

Sort rookies into five buckets by preseason passer rating. The bottom four are statistically indistinguishable from one another. A terrible preseason and a mediocre one and a decent one all produce the same career, once you know where the guy was drafted.

Only the top 20% separates. That group beats its draft-slot expectation by 12 percentile points. In concrete terms: 83% of top-quintile preseason rookies reached 200+ career pass attempts, against 59% of everyone else.

So "preseason is meaningless" is wrong, and "preseason is predictive" is wrong. The truth is narrower and more useful: a genuinely elite preseason is informative. Everything below elite is noise.

Which is why the anecdotes cut both ways

Philip Rivers posted a 46.4 rating as a rookie in 2004, one of the worst preseasons in the dataset, and built a 150-AV career. Aaron Rodgers threw for a 53.0 and went to Canton. Matthew Stafford, 52.8. Eli Manning, 53.7.

Going the other way: Colt Brennan led all rookies in 2008 at 109.9 with three touchdowns and no interceptions, and never threw a regular-season NFL pass. Brett Hundley put up a 129.6 in 2015 and finished with 5 AV. Zach Mettenberger, 97.6 and out.

Both lists are real. That's what a +0.21 correlation looks like from the inside: a real tilt, buried under enormous variance.

And no, we didn't fish for it

Every one of these numbers depends on choices: which efficiency metric, which career outcome, what minimum attempt threshold, which years, whether to include undrafted guys. So we ran all 288 defensible combinations.

Every single one is positive. Median +0.23, range +0.08 to +0.37. The sign never flips. Whatever else this is, it isn't an artifact of the specification we happened to like.

---

Now: Carson Beck

Which brings us back to the game that started this.

The opener

Beck went 15-of-19, 188 yards, 1 TD, 0 INT against Carolina, a 125.4 passer rating on 10.3 adjusted yards per attempt.

Against 162 rookie first preseason games since 2003, that's the 96th percentile. Six rookies in 23 years opened better.

Where it projects

A first game doesn't hold its level perfectly. Across 138 rookies, opening rating carries about half the signal of the eventual full preseason. Run Beck's 125.4 through that relationship and it projects to a 99 full-preseason rating, with a realistic band of 80 to 117.

The number that matters: 99 clears the top-quintile line of 97. The projection puts him, narrowly but genuinely, in the only group in this entire dataset that outperforms its draft slot.

The call

We'll say it plainly: Carson Beck becomes a multi-year NFL starter. 81%.

A typical quarterback taken at pick 65 reaches 200+ career pass attempts about 73% of the time. Beck's opener moves that to 81% (95% CI 59–92%).

And the historical group he just joined is unambiguous. All six rookies since 2003 who opened with a 115+ rating on 12+ attempts became NFL starters. Dak Prescott, Andrew Luck, Baker Mayfield, Jimmy Garoppolo, Bruce Gradkowski, Brady Quinn. Six for six on reaching 200 career attempts, with a median career AV of 60 against 10 for every other rookie.

What his career probably looks like

The tightest comparison set is quarterbacks taken on Day 2 (picks 40 to 120) who then posted a top-30% rookie preseason. Twelve of them since 2003:

92% reached 200+ career attempts. A third cleared 40 career AV. The median finished at 19, and the ceiling of the group is Russell Wilson.

That's the honest shape of the bet. The realistic Beck outcome is the Garoppolo / Brissett / Schaub band: a genuine NFL starter somewhere in the 40 to 70 AV range, with Russell Wilson sitting there at 140 as the reminder of what the right tail of this group looks like.

---

What would change our mind

A bold call should say what would break it, so:

• If Beck's remaining preseason games come in under a 97 rating, he drops out of the top quintile and the whole case collapses back to his pick-65 baseline of 73%.
• The 81% has a 59–92% confidence interval. The low end of that range is a meaningfully different world from the high end.
• Six comparables is six comparables. That group contains both Dak Prescott and Brady Quinn, and the gap between the hot-opener group and everyone else doesn't clear conventional significance on its own. It's supporting evidence, not the load-bearing wall.
• Beck is a 2026 rookie. Nothing about him is in the training data. This is an out-of-sample forecast, which is the only kind worth making, and the only kind that can be wrong.

We'll re-run it after every preseason game.

---

Method: preseason passing lines reconstructed from NFL play-by-play stat IDs and validated 236/236 against ESPN box scores. Careers and draft slots from nflverse and Pro-Football-Reference. Career Approximate Value is Pro-Football-Reference's weighted career AV. All correlations are Spearman rank correlations; the career axis has a large exact-zero mass, so Pearson is not appropriate.