Is my model any good? I have Kristoffer Reitan ranked 97th, and he just won Truist by two shots.

We’re building a model specifically for the U.S. Open at Shinnecock Hills, but this miss deserves an honest accounting going into the PGA Championship at Aronimink. We’re treating the PGA as dry run for Shinnecock, so course set up and other features won’t be live for this tournament. However, the features that caused this miss will have a say in the U.S. Open story too, so let’s introduce them.

I have a bad feeling about this Chewie

The reality is the model didn’t miss Reitan’s recent form. His form residual over the last eight events was +4.07, the highest number in the entire field. First by a wide margin. The model even predicted he’d massively outperform his historical baseline. The problem is his historical baseline is terrible.

Reitan has 13 events of event history, and a lot of those early rounds were….. rough. U.S. Open puns are already flowing. His skill prior of -3.965 is catastrophically negative. When you add +4.0 to -3.965, I can do this math, you get a predicted total of just over Flounder’s GPA of 0.0. Funny how that Flounder guy keeps coming up. This is how a player the model sees as coming into the tournament in great form ends up ranked 97th. That’s a big albatross to hang on someone’s neck. I’ll be here all week.

The model is actually working, but some of the inputs are problematic. A short-term fix would be a post-prediction adjustment that would discount the prior slightly in favor of recent form. With a slight tweak, I can move Reitan from 97th to 25th, which feels realistic. The more principled solution is a prior that decays gracefully toward recent results as a function of recency and events played. Not going to happen by Thursday, but it’s on the roadmap for Shinnecock.

The Young and the Restless

Before we get to the PGA, it’s also worth pointing out what happened to Cameron Young on Sunday. The model had him as the hottest player by form among anyone with real a sample size. 108 events of history, with a form mean of +1.97 on a positive and accelerating trend. He finished T-10 at -9, which included a rough Sunday score of +3.

T-10 at a $20 million signature event is not a miss. But the clean “hottest form wins” story didn’t hold up over 72 holes, and glossing over that to protect the narrative would be the exact thing this brand exists to call out when other people do it. The form signal was real. The result was complicated. Both things are true.

Some names to look out for at the PGA

For the players the model can actually evaluate with some confidence:

Ludvig Aberg is the cleanest case. Sixty-five events of history, so the prior is real if still developing. Form mean of +1.34, and the steepest positive slope in the legitimate-sample portion of the full field. He’s not just playing well; he’s still improving week over week. The market is offering him somewhere around +2000, which is where players land when the ball-striking is undeniable but the wins haven’t come. Four top-five finishes in five events is not a coincidence.

Tyrrell Hatton is the name that doesn’t get enough airtime in this conversation. One hundred twenty-nine events of history, form mean of +1.47, positive slope of +0.38. The prior is slightly negative (-0.18), meaning his historical baseline is fractionally below the field mean, he’s not a career elite, but he’s playing well above himself right now. The combination of real sample size and genuine hot-hand signal is exactly what you want to see.

The upper right quadrant below shows who’s hot and getting hotter….

Not quite ready for Vegas

Don’t take all of this to Polymarket yet. The model doesn’t know what Aronimink looks like and how it will play. Course fit is on the roadmap, but not there yet. Also, we don’t know when Scheffler will put together 4-full rounds of Scottie Scheffler golf and blow away the field.

What the model can say: Aberg and Hatton are carrying genuine form into Thursday. The Reitan caveat stands too. The model missed him once. It might miss him again.

Model output based on rolling 8-event strokes gained residuals. Skill prior anchored to 2025 year-end DataGolf values — in-season improvement not yet captured. Course-specific fit not modeled.

One of the most famous comebacks from a 3-1 deficit, the Cleveland Cavaliers came all the way back against the Golden State Warriors to win the 2016 NBA Title. It was the stark reminder that when playing against Draymond Green, “You’ve got to always protect the McNuggets”, Cheeseburger, The Longest Yard.

This year the Eastern Conference is providing another 3-1 deficit experiment. Two teams came into Round 2 under the same circumstances. Both needed to pull themselves out of a 3-1 hole to win a series that appeared lost. Same narrative. Same emotional momentum. Same “this team found something” energy in the media.

Five games into the second round, and those two teams could not look more different. The Detroit Pistons are up 2-0 on Cleveland. The Philadelphia 76ers are down 0-3 to New York, with the Knicks winning Game 1 by 39 points. There may be only one number needed to explain this variance: the seed.

Not every 3-1 Comeback is the same

Detroit was the 1-seed. They came back from 3-1 against an 8-seed Magic team. The comeback included a 24-point comeback win in Game 6 and a 22-point win in Game 7. They roared into Round 2 looking every bit the best team in the East during the last 3 games of the previous series. When the Pistons’ backs were against the wall, they rallied and showed why they were a 1-seed.

The Sixers were the 7-seed. They beat a Celtics team missing Jayson Tatum in Game 7, survived three elimination games, and showed genuine toughness. But surviving isn’t the same as being good. The 76ers came into Round 2 as a low seed riding emotional momentum, but also dealing with near constant uncertainty around the health of Embiid. Emotion can only take you so far, especially when you find yourself up against a Knicks team that has to get to the Finals to justify firing their coach last season.

When 1 + 1 might equal 3

No 7-seed has ever reached the NBA Finals. Since the league expanded to a 16-team bracket in 1984, 7-seeds are 2-5 in Round 2 heading into the 76ers series. The two previous teams that actually won also greatly benefited from drawing 6 seeds as their Round 2 opponents. History is definitely not on the 76ers side. Even if it is a glass ceiling for the 7-seed, the 76ers were already pressing against it before a single game was played.

The 3-1 comeback adds another layer. Not because 3-1 comeback teams are cursed, the Pistons are actively disproving that, but because of what it signals about where you’re starting from. Teams that go down 3-1 and survive by winning three straight, typically playing at a level beyond what their talent should allow. Regression to the mean is real, and oftentimes, you don’t know when it’s going to happen.

So 7-seeds advancing beyond the second round are rare. Overcoming 3-1 deficits are also rare, only 15 in NBA history. This is what statisticians call additive effects: two independent risk factors that each reduce your odds, and when they’re both present, they compound. The 76ers combine both factors, simultaneously, with no historical comparison. That’s right no 7-seed has ever come back from a 3-1 deficit before the 76ers. So, this is like the “double-rainbow” of the NBA playoffs. “What does this mean? It’s so bright, it’s so vivid”, viral double-rainbow guy. It might just mean a 4-0 sweep for the Knicks.

They are who we thought they were…

The Pistons are showing us that a 3-1 comeback doesn’t break a team. Detroit looks fine. Comfortable, even.

The 76ers are showing us that a 7-seed ceiling is real regardless of how you got there. Three wins in a row against Boston doesn’t change what the Sixers are. It just delayed the moment where the bracket made it obvious.

Two comebacks. Two teams. One is a 1-seed that briefly looked shaky. One is a 7-seed that briefly looked like more than it is.

The East isn’t a mystery. It’s a controlled experiment. And the results are in.

Data via Basketball Reference and NBA.com.

Two teams. Same issue. Different result. Wait until Luka gets back. Wait until Ant returns. Then these series will get interesting. By Tuesday night, both teams had their first data point. They went in opposite directions.

I missed some time, but it only felt like five minutes to me…

Anthony Edwards came back from the Quantum realm, I mean his hyper-extended knee, for Game 1 in San Antonio. Despite a heavy brace, two weeks of rust, and being on the road against the team with arguably the most dominant player in these playoffs, the Wolves won 104-102. Credit where it’s due: that’s exactly what the “just wait” crowd was promising. Ant returned, Minnesota was competitive, and they stole home court from a 2-seed. Score one for the narrative.

Despite the loss in Game 1, the Spurs are still the favorites to win this series. One road win is Real, but it is not a series. Donte DiVincenzo still has a torn Achilles, so his shooting will be missed. And just like coming back from the Quantum Realm everything looks just a little bit bigger. Like looking across the court at a 7’5” Victor Wembanyama posting 21 points, 8.8 rebounds, and 4 blocks per game, and an astonishing 12 blocks in Game 1!

I guess we’ll have to wait and see on this one.

Who is the Lakers’ star in their prime again?

Luka Dončić did not play in Game 1. OKC won 108-90. An 18-point blowout. The team that was supposed to need Luka to make it a series got handled by a Thunder squad missing its own co-star with a hamstring issue. OKC went 4-0 against the Lakers in the regular season. With Luka. The argument for LA was always thin; it just felt less thin when you could point to a healthy version of their best player and say “imagine when he comes back.”

The Lakers are a team getting outclassed, and the math on Luka’s timeline, Grade 2 hamstring, five weeks out, no full-speed workouts, has not improved. The Lakers are now in a hole that Luka would have to dig them out of cold, on limited conditioning, against the best young roster in the West. The series overwhelmingly favors the Thunder, so it’s not a swing series waiting for a star to return.

Sometimes the injury return is the story. Sometimes the team was already losing.

A Tale of Two Teams

Twenty-four hours of playoffs gave us one confirmation and one refutation of the same narrative. That’s not a coincidence; it’s the honest answer to what “just wait” was always covering up.

Ant’s return mattered because he was actually ready to play and because his team was built to contend even before he went down. The Wolves beat Denver in three straight to close the first round. They have infrastructure. Ant returning makes a real team better.

Luka’s return, whenever it happens, would be making a team that just got blown out by 18 into something. Maybe. At some level of conditioning that nobody can project from the outside. Against a team that has beaten them five straight times dating back to the regular season.

“Just wait until he comes back” isn’t wrong as a concept. It’s wrong as a blank check. The return matters. So does what you’re returning to.

Data via Basketball Reference and NBA.com.

I did not see an article on Nick Nurse on my Bingo card this week, but sometimes a stat or a graphic just kind of forces you into it. The power of the Graphic compels you! Eighteen months of Philly columns wanted Nick Nurse fired. 24-58 last year. The team set the all-time NBA record for most unique starting lineups in a season at 54. By Christmas, “this hire was a mistake” was the consensus take.

Then this April happened. Lost Game 1 by 32. Lost Game 4 by 32. Down 3-1 with Joel Embiid recovering from an emergency appendectomy and a rookie playing his first month of playoff basketball. Things were not going well. Then he won Game 5 in Boston by 16, won Game 6 at home wire-to-wire, and forced a Game 7 in a series that was supposed to be, at a minimum, a gentleman’s sweep.

I don’t know whether to fire you or buy you a drink

The city of brotherly love is not for the faint of heart. Nurse’s three-year regular-season record with the Sixers is 91-108. Not great. Most of the Ls were driven by the collapse last year, which was historically bad. His only previous Sixers playoff appearance was a six-game first-round loss to the Knicks in 2024. Embiid availability has been a moving target, but every Sixers coach in the last decade has had access to that excuse. If you stop reading the data there, you fire him. Plenty of people have already written that column.

Ok, but why did you write this?

Nurse’s career playoff record heading into Game 7 is 29-21. That’s a .580 win percentage in the playoffs, with an NBA championship (Toronto, 2019), a Coach of the Year award (2020), and whatever this current run becomes. The most famous “coaching” receipt is the box-and-1 on Steph Curry in the 2019 Finals — arguably the most consequential mid-series adjustment of the last 25 years.

Here is the payoff. Stack his record against every Coach of the Year winner since 2000 with at least 20 playoff games:

The dashed line is the break-even where the coach’s playoff win percentage matches their regular season win percentage. Most of the COY club is below it: Doc Rivers, Tom Thibodeau, George Karl, Mike Brown, all good coaches with playoff win percentages lower than regular season numbers. The 82-game grind built the reputation, but winning on a random Tuesday night back-to-back is a different proposition than winning a 7-game series when you can’t hide who you really are.

Nurse is the red dot. Above the line by 2.7 points. Mark Daigneault, fresh off a title with the youngest roster in basketball, is setting the pace. But here’s who else is on Nurse’s side of the line with real mileage: Steve Kerr. Four championships, the modern gold standard for a coach worth his contract every April.

Once you exclude the small samples, Kerr and Nurse are essentially alone in the “demonstrably better in the playoffs than the regular season” club. Even Pop isn’t on this guest list.

Why the postseason hits different

The regular season and the playoffs are not the same sport. They look like the same sport. They are not. They’re played differently. They’re officiated differently.

Regular season: 82 games, schedule load, rotation depth, you live with what your roster gives you most nights. A coach whose identity is “make tactical adjustments and shrink the rotation when it matters” is going to look ordinary across that scenario, because it’s not rewarded.

Playoffs: the same opponent for seven games over two weeks. Adjustment over adjustment over adjustment. The coach who can change his offense and defense between Tuesday and Thursday and have his rookie execute it cleanly wins games his roster shouldn’t.

A coach optimized for the second scenario looks ordinary in the first. That’s not a bug. That’s a feature.

A Tree with Multiple Branches

If the Sixers win tonight, Nurse will have engineered only the second instance in NBA history of a team coming back from two 30-point losses to win a playoff series.

If the Sixers lose, they will have pushed a 56-win team to a Game 7 in a series that was 3-1 with their biggest star recovering from emergency surgery. If this is the floor of what Nurse can do under those conditions, what’s the ceiling with a healthy roster and a normal offseason?

Whatever the result, regular-season Nurse and playoff Nurse aren’t the same coach. Wouldn’t we call a player clutch if they did the same?

Data via Basketball Reference and NBA.com.

The Cavaliers entered the playoffs with Donovan Mitchell, Evan Mobley, and James Harden, a lineup assembled through a combination of draft capital, player development, and a February deadline swing for a future Hall of Famer. The Raptors entered as the five-seed in the middle of a rebuild that isn’t quite finished yet. Somehow this series is way more competitive than expected.

That gap between expectation and reality is what the superteam tax looks like when the bill arrives.

What did it Cost? Everything.

OK, settle down Thanos. In 2022, Cleveland traded a haul of players and first-round picks to Utah to acquire Donovan Mitchell. Then in February, they traded Darius Garland, a young All-Star-caliber guard, and a second-round pick to the Clippers for Harden.

The logic was somewhat defensible. Harden was playing well in Los Angeles. Garland was good, but injury prone. The move was designed to go deeper in the playoffs, not to survive a first-round series against a team rebuilding around Scottie Barnes. Admittedly, I watched Barnes play at FSU, and I really didn’t get it, but he has blossomed into a force this year.

Don’t tell my wife, but I am wrong sometimes.

But here’s what superteam math almost always misses: you’re not just buying a player. You’re trading future flexibility for present upside. Unfortunately, that present upside often results in your championship aspirations turning into dust.

I did not see that coming…

Here is where the Harden logic being defensible falls apart, and falls apart fast. No denying, James Harden is one of the greatest offensive players in NBA history. But when you’re making an acquisition to go deeper in the playoffs, maybe you don’t choose a guy who has one of the most well-documented performance drags between the playoffs and the regular season in the NBA.

The chart above isn’t just indicting Harden the player, it’s indicting the premise that what you’re getting in February is the same as what you’re getting in May. Toronto figured out quickly that trapping Harden and Mitchell simultaneously and daring the rest of the roster to beat them was a viable strategy, because the rest of the roster is exactly what Cleveland traded away to get Harden in the first place.

That’s the circular logic at the center of the superteam tax. You acquire a star to compensate for roster depth. Acquiring the star depletes roster depth. The depth you traded away is precisely what the other team exploits in the playoffs.

OK got it, so don’t get the player who wore a fat suit?

Two teams, same bill

It’s not quite that simple. If Cleveland is the slow-motion version of this story, Phoenix is what it looks like at full speed. The Suns traded a massive haul for Kevin Durant and Bradley Beal. Young players. Multiple first-round picks. The kind of assets that build contenders over time . The Goal was a championship. The result: both players are no longer on the roster, the Suns were an eight-seed, and they just got swept by Oklahoma City.

OKC, for the record, built their roster with assets. Draft picks. Patient development. The exact kind of moves that Phoenix gave away. The second best team in the West, San Antonio, didn’t trade for their cornerstone, they waited for him. No assets surrendered, no future mortgaged, no elaborate sign-and-trades. The Spurs did the unglamorous thing: they were bad for a couple of years, got the right lottery ball, and let a generational talent develop, while putting pieces around him that made sense. It’s the direct opposite of what Cleveland and Phoenix tried to do.

OKC and the Spurs look like the class of the West. Both built through the draft. Neither mortgaged the future to rent a ceiling. The bracket this year is practically a controlled experiment: on one side you have teams that paid the superteam tax; on the other, you have the teams that collected on it.

Even a broken clock…

Superteam construction isn’t always wrong. Miami’s Big Three worked. Golden State is complicated but their dynasty was draft-built first and augmented second. The approach can work, it just works far less often than the teams attempting it seem to believe. The failures are often brutal because the teams that try it are left without the assets to rebuild. Hello Milwaukee!

Cleveland’s situation isn’t hopeless. They can still win this series. But the fact that it’s even a series, the fact that a front office that spent this much is watching a 2-2 scoreboard against Toronto, is the data point. The gap between what was purchased and what’s been delivered is exactly the gap the superteam tax predicts.

While you can buy a ceiling, you can’t buy certainty. And in the first round, against a team nobody was worried about two weeks ago, that distinction matters.

Data via Basketball Reference.

I know, I know… Not another super technical article. I promise this will be the last one for awhile, and we can go back to the NBA playoffs, the Majors, and anything else fun that comes up. But first, you gotta meet this guy!

Hello World, Earnest

In the last article, we talked alot about how “we” built the model to predict boom/bust probabilities in the NFL Draft. I deliberately used the word “we” a lot in that article, and it’s time I tell you why. Meet Earnest!

He is an AI automation agent, built on OpenClaw, running on the aforementioned Mac Mini in the corner. He has one job, extend Merrittocracy’s reach without replacing my voice. “It’s what he does, it’s all he does.” But unlike the Terminator, he actually can be reasoned with. Whether he’s monitoring for new Substack articles, drafting X-posts in my voice, or surfacing reply opportunities, Earnest routes everything back to me through Telegram for approval before anything leaves the friendly confines of Merrittocracy.

We actually go back and forth quite a bit, but he does think all of my ideas are brilliant. Of course he does. The approval loop is the whole point. Earnest doesn’t publish. He proposes. Every draft goes through Telegram, every post requires an explicit yes, every irreversible action has a “human in the loop”. He also files his own journal entries when something significant happens. This would normally be great, but man, my ego is taking a beating seeing all of the mistakes I’ve made along the way.

Earnest went live for the first time during draft week. Getting him there was its own project, with its own set of challenges to overcome. He exists, he works, he has a soul, and he’s a character in this story, not just a tool in the stack.

Wait, did you say soul? Yes I did, and it’s one of the cooler things about building an AI with OpenClaw. During the onboarding process, you develop a set of instructions that end up becoming the agent’s “soul”, and the file is called SOUL.md. It’s why he responds like this…

So wait, is this all just AI?

In a word, NO! I’ve always considered myself a better editor than initiator. So, yes every draft article does start as a back and forth with either Earnest of Claude, by Antrhopic. But, that is not what lands in your inbox. My writing process is “unique”. I guess this is the part of the article where my nerd takes over. The draft that I create with AI drops into a content folder in a version control software system. This system tracks every change I make with the draft, and then it is posted online.

What is displayed in the software is called the diff and the diff is the receipt. Anyone who wants to ask “is this just AI?” can go look at the answer on the version control site and look at my history of changes for each article. The gap between what AI produces and what I publish is where the editorial judgment lives, and that gap is the whole argument for why AI as a force multiplier works. It’s not replacing my voice. It’s handling the scaffolding so my voice can focus on what matters.

No AI would ever title an article The Big Arch vs. The Mendoza Line or F** Them Picks*.

The Infrastructure That Didn’t Survive First Contact

Actually, this entire section is not going to survive first contact. Most people are already not forgiving me for the last two articles, so I’m not going to put you through it again. If you’re technically inclined and want to check out how all this came together from a nerd’s eye view, you can find everything on the two sites below. Everything is public.

• github.com/merrittocratic/autopilot

• github.com/merrittocratic/OpenClaw-Ops

Just in case you’re wondering if AI is ready to rule the world, I can honestly say, we’re not quite there yet. If you didn’t hear, Cooper Flagg won the NBA ROTY award, and I had written an article last week about how I felt he deserved it over Konn Knueppel. The data supported it, which is kind of the point of Merrittocracy. Earnest is supposed to look for X-posts related to the content I’ve written, so I would have thought when Shams Charania announced the award winner, Earnest should have notified me ASAP. That’s not what happened. Apparently, Earnest thought my Are We Being Konn-ed article was about Konnor Griffin. Who is that you ask? Apparently, he is a MLB prospect going to the Pirates. Definitely not the all-time rookie leader in 3-pointers made.

Draft Night

I want to be honest about what it felt like on draft night after a month of not only building a predictive model, but also building an AI agent to amplify my results.

I built a script to quickly score picks as they became public, and… It worked! Picks came across the ticker, I typed pick(“Kenyon Sadiq”), the pick importance waterfall populated, the boom/bust probability populated, and Earnest had a draft thread in Telegram ready to post to Adam Schefter’s and Mel Kiper’s X timeline within seconds. The infrastructure that wasn’t even an idea in late-March became something very real on a Thursday night in April.

I used Sadiq as an example here, because the organizational tax landed exactly where the model said it would. Players like Sadiq, Cooper, Iheanachor, Rutledge all carried the development penalties we’d documented, and most analysts talk about but have never quantified. The model didn’t just run. It said something.

So What’s Next

The 2027 model has a long wishlist: conference tiers, birth dates sourced at the start of the season instead of patched on draft eve, proper player ID joining instead of name matching across three databases. Structural fixes that would have made this cycle cleaner.

The NBA Playoffs are already generating buzz. The PGA Championship and U.S. Open are just around the corner. I’ll keep posting, and in the background I’ll be building a U.S. Open model. Who knows, maybe I’ll even write a WNBA article. Just kidding, that’s in the same bucket as Hockey. Not gonna happen.

No matter what I write about, somewhere on the Mac Mini in the corner of my office, Earnest will be running his heartbeat check, monitoring for new content, filing journal entries, and routing around the next blocked feature before I ever ask him to.

Earnest Lives! github.com/merrittocratic/autopilot.

In the beginning there was Homebrew.

For the non-nerds, Homebrew is a package manager, a tool that installs other tools. It’s the foundation of any modern Mac development environment, it’s the thing you put in before everything else. But what makes Homebrew worth talking about isn’t just what it does. It’s how it “thinks” about what it does.

Homebrew is built entirely around a craft brewing metaphor. You don’t add a software source, you tap a repository. Yes, like tapping a keg. Nerds drink beer too, you know. You don’t install packages, you install formulae. Graphical Interface applications live in casks. Everything that gets installed lives in the Cellar. The whole system is designed as if a brewmaster sat down and asked: what if the thing developers do every day felt a little more like making something with your hands?

That framing turned out to be exactly right for this project.

One nerd. A $500 Mac Mini in the corner of a home office. Open-source tools stacked on open-source tools. Formulae tapped from repos across the internet, assembled into something that didn’t exist four weeks ago. We brewed a sports analytics operation from scratch: models, pipeline, automation agent, content engine, and it was finished just in time for the 2026 NFL Draft.

Merrittocracy is about receipts, and with the draft now over, it’s time to give everyone a view under the hood of how this was all cobbled together. Here’s what one month of building looks like from the outside:

1 month. 4 repos. 12 articles. 15,000 words. 8 position-specific machine learning models. 1 automation agent. 1 sports analytics brand that didn’t exist on March 29th.

We tapped the repos. We ran the formulae. We poured what came out.

Time to “Pour one out for the homies.”

If you can be anyone, be yourself… Unless you can be Batman, then be Batman

Lucius Fox: “I analyzed your blood, isolating the receptor compounds and the protein-based catalyst.”

Bruce Wayne: “Am I meant to understand any of that?”

Lucius Fox: “Not at all. I just wanted you to know how hard it was.”

That’s the honest version of a building-in-public article. I’m not going to make you a machine learning engineer. Not even going to try. What I want you to walk away with is a feel for the actual terrain, why it took a month to build all of this, why plans that looked airtight on paper didn’t survive the first punch in the mouth, and why the final product is built differently than the version I originally designed.

The technical details are receipts. The story is the journey.

The Decisions That Changed the Model

Building any model involves a long series of choices that look minor until you see what they were actually hiding. There are three that I definitely want to highlight, as they straddle both real football categorization questions and also impact model reliability.

Before getting into what broke, it helps to know what we were actually trying to predict. The outcome variable is Career Approximate Value, a catch-all production metric from Pro Football Reference, measured over a player’s first four NFL seasons and adjusted for where they were drafted. Beat your draft slot expectations by enough and you’re a boom. Miss them badly enough and you’re a bust. Everything in between is a J.A.G. Just A Guy…

The DB Split. The original design grouped Cornerbacks and Safeties together as “defensive backs.” Reasonable football intuition, same side of the ball, similar evaluation criteria, combined sample size is comfortable. The problem, CB booms 4% of the time. Safety booms 12% of the time. Very different outcome distributions, and their results began to cancel each other out. Splitting them was the right call, and it was the move that sharpened the earlier Caleb Downs piece considerably. Once Safety stands on its own, the positional value argument gets a lot stronger. More importantly, you also start to realize why a CB with an injury history that was widely expected to go in the first round drops to the Raiders on day 3.

The Imputation Effect. When you have missing data, one of the most common moves is to fill it with the median value and move on. We originally did that for college defensive statistics, where the data source doesn’t cover seasons reliably before 2012. For Linebackers, roughly 65% of the training data predates that cutoff, so all of those LBs got the median fill. The remaining 35% had real, measured statistics. Our best performing models saw that behavior immediately, and all of a sudden, interceptions for an off-ball linebacker became more important than tackle production. Because “median fill versus something else” is a dead-reliable indicator of whether a player was drafted before or after 2012. The model was predicting era, not the real outcome. We removed the imputation, and let the model handle the missing values natively, it didn’t need us to fill in blanks, it’s perfectly capable of reading on its own.

WR/TE Can’t Share a Scale. A tight end with 60 receptions in a college season is an elite, alpha-level weapon. A wide receiver with 60 receptions is fairly mid. But when both groups compete for rank in the same percentile table, every tight end looks like a below-average receiver and every wide receiver looks like a monster, because WRs catch far more passes by design. Both signals get distorted, and the model falls back on draft position as the primary predictor. This defeats the purpose of having college production features at all. The fix was straightforward: rank each position within itself. WRs against WRs. TEs against TEs. The solution was obvious the moment I stopped thinking like a data engineer and started thinking like a scout.

This is why scouts and GMs still matter. None of the issues above produced errors. In fact, they produced valid-looking outputs, that just so happened to be wrong. It’s why we still need some Man-altyics around here, because a model won’t wave a flag to say when it’s confused or picking the noise instead of the signal. Does it make sense where a QB ranks compared to his peers in receptions? Certainly, a novel feature to add to a model, but if that one is leading the pack in the prediction, probably time to ask some questions.

The Pick Ladder

Some design decisions were actually proactive rather than corrective. And they worked!

The model doesn’t score a prospect at a single projected draft pick. It scores him across a range of picks, a ladder of positions from 1 to 257, and shows how boom and bust probability shifts across that range. “Simpson at pick 20: 18-23% boom. Simpson at pick 45: still 21-27% boom.” Honest uncertainty ranges instead of false precision, and the model does not rely on mock drafts for accuracy, which is genuinely unreliable past round two.

The design question was what intervals to use on that ladder. The obvious answer is clean steps of 5 or 10: picks 1, 5, 10, 15, 20, 25, 30. Tidy. Easy to explain, but also very wrong.

Round numbers and pick numbers are as much psychological categories as actual positional values. “Top-10 pick” carries meaning that has nothing to do with the mathematical distance between picks 9 and 11. The ladder uses increments of 4 from pick 1 until the end of round 2, then increments of 10 through round 7, threading past any anchor numbers without hitting them.

Why did they change Math? Math is Math.

But in the draft, I think we would all agree, the value difference between picks 1 and 5 is not the same as the value difference between picks 25 and 29. You probably thought you’d never hear about logarithms again, but unfortunately or fortunately, the historical production curve for draft picks follows a steep log function at the top and flattens dramatically by the end of the first round. Moving from pick 1 to pick 5 is a significant devaluation. Moving from pick 25 to pick 29 barely registers.

Equal distances. Completely different value gaps. Nobody notices the ladder design in the outputs. Nerding is a behind the scenes sport.

What’s in a Name?

The model joins data from multiple sources, primarily: nflreadr for combine measurements, cfbfastR for college statistics, and our mock board CSVs. None of them use the same name for anything! Again, no crashes, no pipeline failures. Just panic, when you notice you’re missing a prediction for a top Defensive Line prospect when the Jerry Jones does Jerry Jones things. He goes out and drafts Malachi Lawrence out of UCF or “Central Florida” or the University of Central Florida depending on the source.

Did you know that Alex Styles was a really good LB out of Ohio St. or OSU or The Ohio State University? Me neither! Because, everyone else knows him as Sonny. Luckily this was found before the draft during testing when the Sonny Styles prediction that was modeled came back based on 0 combine data. His “legendary”, “cyborg” workout should have helped his prediction. It was helping someone, alright. Alex, not Sonny.

This is the unglamorous reality of building a pipeline across sources that were never designed to talk to each other. The sophistication of the model is completely irrelevant if the names don’t match.

Again, “I just wanted you to know how hard it was.”

Three Algorithms Entered. One Algorithm Leaves.

While it wasn’t quite the Thunderdome, we did run three fundamentally different algorithms against the same data, the same cross-validation folds, and the same outcome variable. Here’s how it went.

XGBoost is the gradient boosted tree that the analytics community has been using for fifteen years. Tuned across fifty hyperparameter combinations per position group. The incumbent.

TabPFN is a foundation model published in Nature, pre-trained on millions of synthetic tabular datasets. It makes predictions in a single forward pass, no training on your data, just pattern recognition from a model that’s seen more tables than any individual problem will ever produce. The challenger with the pedigree.

TabNet is attention-based, deep learning built specifically for tabular data, with the ability to show you what the model focused on for each individual player. The technology demo.

XGBoost won all eight position groups. TabPFN managed a marginal improvement on two. TabNet didn’t beat the baseline, just predicting the mean for every player, on a single group. On Offensive Linemen, TabNet posted an RMSE of 2.08 against a null model of 1.0. Twice as confused as doing nothing.

Oh yeah, and it took FOREVER to run, RUN_TABNET <- FALSE is now a permanent flag in the codebase.

Here’s the honest explanation for why this was predictable: the academic literature on tabular machine learning is consistent on this point. Tree models dominate at small-to-medium data sizes with mixed feature types and low signal-to-noise ratios. Our dataset is all three, between 162 and 545 players per position group, a mix of numeric and categorical features, and an outcome variable that’s inherently noisy because pre-draft data can’t see coaching quality, injury luck, or scheme fit.

Next year, I might try creating simulated data for TabNet and try to get up to 2-3k per position group (probably still not enough). Unfortunately, for this year, the fancy model would have lost to a spreadsheet.

That’s not a failure. That’s a result and a storyline.

The Feature Nobody Planned For

The original model design had combine measurables, college production, draft position, age, and the college program pipeline. The drafting organization was not in scope.

It got added mid-cycle, driven by a content need. The “Van Isn’t the Variable” piece on team development quality required building a rolling 10-year pick-adjusted AV residual per franchise. We wanted to know how much better or worse does each organization do at developing players relative to what their draft slots historically produce? The feature went into the model to support one article.

On draft night, it became the story.

Kenyon Sadiq landed with the Jets. His predicted z-score swung from +0.169 pre-draft to -0.08 post-pick. The entire swing, a quarter of a standard deviation, came from one feature: the Jets’ player retention rate, a proxy for organizational development continuity. Omar Cooper Jr. took the same penalty at pick 30. Two players, same organization, same feature, same discount applied before either of them had attended a single practice. It makes sense, the J-E-T-S do not have a great history of player development, and I don’t know that the current coaching staff can turn it around.

Max Iheanachor to Pittsburgh: the Steelers have produced zero first-round offensive tackle booms in fifteen years of training data. Keylan Rutledge to Houston: same feature, same value. Two offensive linemen. Two organizations with empty first-round OL boom columns in the historical record.

The model doesn’t hate any of these players. It has documented, specific concerns about where they landed. Some destinations amplify talent. Some discount it. That signal was sitting in fifteen years of data. We found it by accident while writing a different article.

The NFL Draft sells hope every year. Unfortunately, most of that hope is wrapped up in the player and not the organizations drafting them.

Let’s wrap it up people

XGBoost wins. A tuned gradient boosted tree trained on 3,750 historical prospects outperformed two transformer-based architectures on every position group. The algorithm the analytics community has been using for fifteen years is still the right tool at this data scale. That’s not a failure of ambition. It’s a result.

Pre-draft prediction has a ceiling. RMSE ranges from 0.948 on quarterbacks to 0.999 on running backs, against a null model of 1.0. The signal is real but modest. Coaching quality, injury luck, and scheme fit explain outcomes that no combine measurement can touch. A model that quantifies what it can’t predict is more trustworthy than one that doesn’t admit the question.

The program pipeline works. Rolling 10-year, position-specific draft outcomes per college program, leave-one-out computed, is a genuine differentiator. Most public models treat college program as a flat categorical. Ohio State WRs are not Ohio State QBs. The 2026 class confirmed the feature adds signal.

The organizational tax is real. Built mid-cycle to support one article. Became the organizing principle of draft night coverage.

That’s the model. That’s what it took to build it, what broke along the way, and what the data found when we stopped arguing with it. But a model sitting on a laptop isn’t a brand. Getting the findings out the door on draft night, in real time, while picks were coming off the board, that required a whole separate operation. One that had its own set of plans that didn’t survive the first uppercut, its own cast of characters, and one agent in particular who has a soul, files his own journal entries, and helps others see what I’ve written.

That’s the next piece.

Model trained on 2006–2020 NFL draft classes using pick-adjusted 4-year Career Approximate Value. Program pipeline features computed using a rolling 10-year leave-one-out window. Organizational development features computed as rolling 10-year pick-adjusted AV residuals by franchise. Data via Pro Football Reference and nflverse. Full methodology at github.com/merrittocratic/nfl-draft-model.

So, when I tell you that what LeBron James is doing right now is genuinely insane, I need you to understand that’s not a hot take from a LeBron stan. That’s a confession from someone who has been on the other side of the argument since the “Kid from Akron” entered the league.

The “kid” is now 41 years old, and averaging 25.3 points, 9.7 rebounds, and 8.7 assists in the 2026 playoffs. The Lakers are up 3-0 on Houston, one game from the second round. And he’s doing it without Luka Dončić, the reigning scoring champion and the guy this team is actually built around. Oh yeah, and Austin Reaves is out too.

LeBron is not the centerpiece of a championship roster, but what he’s doing right now might be even more impressive.

Take Two and Call Me in the Morning

Here’s what I keep coming back to. Pull the top-two stars out of every legitimate contender’s roster and ask yourself honestly whether they would make it out of the first-round.

Take Shai Gilgeous-Alexander and Chet Holmgren off Oklahoma City. Take Victor Wembanyama and De’Aaron Fox off San Antonio. Take Jayson Tatum and Jaylen Brown off Boston. I feel pretty confident saying none of those teams would be up 3-0 in their first-round matchups.

And here’s the one that really lands: take Nikola Jokic and Jamal Murray off Denver. I love Jokic! Many consider him the most skilled offensive player alive. Without Jokic, the Nuggets aren’t a first-round threat. And with him? They’re currently losing to Minnesota in their first-round series.

The Lakers? No Luka. No Reaves. No problem. They’re one win away from the second round. LeGM deserves some love.

A List of One

I want to give you some historical context on 41-year-olds averaging 25 points per game in the NBA playoffs.

That list has one name on it.

The oldest player to even score in a playoff game was Robert Parish, who dropped 2 points for the Chicago Bulls at age 43. Not 2 points per game, 2 points total, as a ceremonial cap piece at the end of a career. That’s the pre-LeBron data point on NBA elders in the postseason.

Look at the plot below. This is the playoff series PPG and Age for every player over the age of 37 in the playoffs near 20 PPG. See a pattern?

It’s almost ALL Lebron!

Lebron’s Game 3 line, 29 points, 13 rebounds, 6 assists, 3 steals, and a block in 45 minutes of overtime, would be a career highlight for most players in the league right now. For LeBron at 41, it was like any other Friday night.

There is no historical comparison to make, because history didn’t think to make room for this.

Put some respect on his name…

I’m not here to relitigate the GOAT debate. MJ is still the GOAT to me and I don’t think that changes. Six rings, six Finals MVPs, never lost in the Finals, that’s a case that doesn’t break.

But “who’s the greatest” and “give this man his credit” are two different conversations, and we’ve spent so much energy on the first one that we’ve been stingy with the second.

LeBron James is doing something in 2026 that nobody has ever done, at an age when most players are either retired, doing analyst work on ESPN, or logging 8 ceremonial minutes a night. He is the guy on a playoff team, winning games in April, and right now he’s doing it with his two best teammates watching from the sideline in street clothes.

That deserves some flowers.

Les Snead made that phrase famous in Los Angeles. For a decade, the Rams’ GM was allergic to draft capital, trading picks, flipping futures, mortgaging tomorrows to chase the ring that finally arrived in 2022. His philosophy was simple: if you have the players to win now, draft picks are just chips on the table. F*** them picks.

Thursday night in Pittsburgh, Snead did something that violated the whole premise. With the 13th overall pick, the Rams selected Ty Simpson. Snead’s old line was about not hoarding capital when you’re good enough to win. Taking Ty Simpson at 13 is something different, it’s spending premium capital on a projection pick for a team with Matthew Stafford, a closing window, and a roster built to compete now.

We wrote an entire article last week explaining why Simpson’s production profile doesn’t survive contact with the data. The short version: since 2010, four quarterbacks were drafted in the first round with fewer than 15 career starts. The only one who worked was Cam Newton, who accounted for 4,330 yards of total offense and 51 touchdowns while winning the Heisman and a national title. Simpson accounted for 3,700 yards of offense and 30 total touchdowns on a team that got bounced in the first round of the playoffs. These are not comps.

Here’s where it gets complicated: our model actually likes Ty Simpson.

Predicted z-score: +0.30. Every major driver is positive, TD percentage percentile, a low interception rate that ranks in the top 12%, the Rams’ organizational residual mean. The player-level production signals the model keys on are genuine. And the Rams, whatever you think of the strategic logic, are one of the few organizations where the development environment doesn’t actively hurt the projection.

So what’s the verdict? The model and the narrative disagree, and the honest version is this: Simpson the player may be fine. Simpson the pick, on this roster, at this cost, is still hard to defend. If Stafford gets hurt and Simpson has to play meaningful snaps in Year 1, nobody will care that the model liked his TD rate. But if Stafford plays out his contract and Simpson develops in the background, this could look shrewd in 2028. The Rams may be the only organization in the league with the roster depth to make this bet without it costing them a season.

The real lesson from Simpson isn’t about the player. It’s about the first thing our model taught us this offseason: where you land matters as much as who you are.

When you’re a Jet…

The New York Jets ended Thursday night with three first-round picks and real optimism for the first time in years. We’ve got some concerns, and they all trace back to the same source. Say it with me New York, J-E-T-S! It’s gonna get real Game of Thrones around here.

Kenyon Sadiq, TE, Oregon (Pick 16): Before last night, our WR/TE model scored Sadiq at +0.169, a legitimate positive, one of the cleaner tight end profiles in a class where the position has real first-round value. We liked him so much, he was a focal point of an article. He’s young, his rushing production percentiles are elite for the position, and the pre-draft projection was genuinely encouraging.

Then the J-E-T-S happened. The post-draft projection: -0.08.

The entire swing, a quarter of a standard deviation, is explained by one feature: the Jets’ two-year player retention rate, which sits at 0.161 in our training data. Low retention means players leave or get cut before the development window closes. It means the organization hasn’t historically turned first-round investments at the position into the outcomes the capital should buy. The model doesn’t hate Sadiq. It hates where he landed.

Omar Cooper Jr., WR, Indiana (Pick 30): Same story, same tax. Cooper was widely mocked in the upper half of Round 1, slipped to 30, and the Jets traded back into the first round to get him. His predicted z: -0.11. The dominant negative driver? team_all_retention_2yr = 0.161. The exact same value as Sadiq.

Two picks. Two players with legitimate profiles. The same organizational feature dragging both projections underwater. The Jets are going to have to develop their way out of a track record the model has priced in. To be clear: both players can outperform their projections. The model isn’t destiny. But for a franchise trying to break a significant playoff drought, spending three first-round picks while carrying a development discount on two of them is not how you want to start the night’s scorecard.

The exception is David Bailey at No. 2, our front seven piece made the organizational case for taking a conventional edge rusher over a chess-piece hybrid, and that logic holds. Bailey’s model output (-0.42 predicted z) comes with its own concern: a 95th percentile sack rate paired with a 43rd percentile overall defensive production score, the same pattern we’ve documented in pass rushers who don’t translate. Hello, Chase Young! But at No. 2, the Jets’ track record with conventional edge defenders grades better than their track record with hybrid linebackers. The Simmons template is institutional memory. Bailey avoids it.

Flounder’s GPA of 0.0 is still bad

The two quietest picks of the first round — and the two that concern the model most as organizational statements — belong to the Steelers and the Texans.

Max Iheanachor, OT, Pittsburgh (Pick 21): Predicted z: -0.01. Essentially flat. The model isn’t worried about Iheanachor the player. It’s worried about one feature that dominates his waterfall above everything else: team_pos_boom_rate = 0.

The Pittsburgh Steelers have never produced a first-round offensive tackle boom in our training window. Not once. The model has seen 15 years of Steelers OL first-rounders and the boom column is empty. The consensus take frames this as a depth and development pick — parked behind Broderick Jones and Troy Fautanu, no pressure, long runway. That framing is generous. An organization with a 0% first-round OL boom rate isn’t a patient developer. It’s an organization that has spent first-round capital at the position and consistently not produced the outcomes that capital should buy.

Keylan Rutledge, IOL, Houston (Pick 26): Predicted z: -0.01. Same feature, same value: team_pos_boom_rate = 0. The Texans have also never produced a first-round Interior OL boom in the training window.

Two interior offensive linemen. Two organizations with zero OL boom production in 15 years of first-round data. Both players are individually neutral. Both picks have an organizational ceiling problem baked into them before the players touch an NFL field.

Rutledge generated genuine buzz at the Senior Bowl. His one-on-one reps were impressive and evaluators loved his quickness for the position. The model isn’t saying he can’t be good. It’s saying the Texans, historically, haven’t been the place where “good at the Senior Bowl” turns into “first-round ROI.”

“Jerry, if you find a suit that fits, you buy it! ….. You don’t look for flaws! (G. Costanza)”

Dallas entered Thursday with a defined strategy, trade up if a target falls, trade down if the board goes sideways, and they managed to execute both, while confusing everyone.

Trading up from 12 to 11 to grab Caleb Downs was correct. Downs was the consensus best player on the board at his position, Ohio State’s safety pipeline is legitimate, and the Cowboys’ secondary needed a genuine playmaker. Good pick. Better price.

Then at 23, the Cowboys took UCF edge rusher Malachi Lawrence. Predicted z: -0.38 — the most concerning model output among any of the picks we’ve covered. This all seems so familiar… 45th percentile overall defensive production, 92nd percentile sacks. We documented this pattern in our front seven piece and we saw it again in David Bailey’s waterfall. Elite sack production without broader defensive impact is a known translation risk. The gap between what Lawrence does against college tackles in favorable downs and what he’ll be asked to do on third-and-7 in the NFL is real.

A new wrinkle, the Cowboys actually tend to nail their drafts! Their team_all_boom_rate of 0.316 is one of the higher values in the league. In this case though, it is actually pulling Lawrence’s prediction down, not up. High boom rate organizations have developed well across positions, and Lawrence’s profile isn’t matching the Cowboys’ historical production pattern. So this pick is going outside of the norm for what their development environment has historically turned into outcomes.

To make matters worse, the Cowboys facilitated a trade with the Eagles that let Philadelphia move from 23 to 20 and take Makai Lemon. So, the Eagles got a Biletnikoff Award winner. The Cowboys got Lawrence. One of those transactions is going to sting in NFC East games for years.

I want Winners!

New York Giants

John Harbaugh entered Pittsburgh with two top-10 picks and a mandate to rebuild. He left having executed the clearest strategy of anyone in the first round.

At No. 5, Arvell Reese had slipped from where most projections had him, legitimate top-3 value, and the Giants took him without overthinking it. Does anyone remember the pass rushing talent that beat the undefeated Patriots in Super Bowl XLII? History is starting to rhyme again. At No. 10, the pick acquired in the Dexter Lawrence trade with Cincinnati, they added Miami offensive tackle Francis Mauigoa. A high-upside edge defender and a starting-caliber tackle, both at value, both in the first round. Our model’s chess-piece caution on Reese is real, the deployment risk matters and Harbaugh’s track record developing defined defensive roles is the primary argument it resolves positively. But as draft nights go, the Giants didn’t beat themselves. That alone puts them near the top of the evening.

Cleveland Browns

Yes Cleveland! The Browns traded down from No. 6 to No. 9 (collecting a third and fifth from Kansas City in the process), landed Spencer Fano, the cleanest OL prospect in the class by multiple evaluations, and then used the Jacksonville pick at No. 24 to add Texas A&M’s KC Concepcion at receiver. Two first-round players at genuine value, one of them funded by prior trade capital. They let the board come to them and pocketed surplus along the way. Are the Browns geniuses, or are they morons who stumbled into the right answer. Time will tell.

Tampa Bay Buccaneers

Rueben Bain Jr. at 15 is the most defensible pass rusher pick in the round. Conventional deployment, clear role, no chess-piece ambiguity. The EDGE bust rate in Round 1 is the highest of any front-seven position, we documented it, but Pick 15 for Bain is better than pick 7 or 8. Last year’s NFC South Champion, the Panthers, are about to be asked: “Do you feel in charge?”

And that’s the Bottom Line!

The consensus take on Round 1 of 2026 is that it was weird, unpredictable, philosophy-driven, and hard to grade in isolation. That’s true. But the model offers a lens the discourse mostly missed: the destination shapes the projection as much as the player.

Sadiq is good. The Jets’ development environment is the variable. Iheanachor and Rutledge may be fine players, but they’re going to organizations that have never turned their position into first-round returns. Simpson’s player-level profile is stronger than his narrative suggested, and the Rams may be one of the few places that can make the math work on a developmental quarterback pick.

The draft doesn’t just select players. It assigns them to development environments with documented track records. Some of those environments amplify talent. Some of them discount it before the player has attended his first practice.

Boom/bust probabilities and predicted z-scores derived from the Merrittocracy model, trained on 2006–2020 NFL draft classes using pick-adjusted 4-year Career Approximate Value. Organizational pipeline features computed using a rolling 10-year window with leave-one-out methodology. Data via Pro Football Reference. Full methodology at github.com/merrittocratic/nfl-draft-model.

The 2026 draft has a receiver narrative. Deep class. Lots of options. Get in early. Carnell Tate, Jordyn Tyson, and Makai Lemon are all projected inside the top 15. Even the people selling the narrative are hedging. Bucky Brooks: “I wouldn’t say it’s elite; I would say it’s really good.” Jordan Reid: “We don’t have those star players. We have a lot of No. 2 wide receivers in this class.” That’s not a ringing endorsement of paying top-12 money. No Jefferson. No Chase. The Model agrees!

Two different asset classes

The Blue Chips:

Carnell Tate is the best of the group, and he’s legitimately good. Predicted z-score of 0.18, 20.5% boom, 11.6% bust. He’s the youngest prospect in the group relative to his position peers, he’s a strong mover in space, and the Ohio State WR pipeline (8 drafted, 15.9 average 4-year AV) is one of the more reliable program signals in our prediction. This combination of youth, athleticism, and a proven development track earns him the highest predicted score of the four.

That said, he’s a quality prospect in a down year, not a generational one. Daniel Jeremiah has him #7 overall, and our prediction doesn’t disagree with the ranking. It just thinks the market is confusing “best available at the position” with “elite.” Those aren’t the same thing.

Jordyn Tyson has the lowest predicted score of the group at 0.04. The largest detractors on his score are his program pipeline: Arizona State’s historical WR outcomes and his projected draft position (inside the top 10 on some boards) make his selection risky. We expect more from higher picks, and Tyson’s profile doesn’t clear the bar. Elite athlete. Real injury history. Our prediction is skeptical, and the spread in where evaluators have him ranked is that skepticism showing up in real time.

Makai Lemon sits in the middle at 0.11. Same problem as Tyson, weak program development for wideouts (USC, thin WR history), the weird mannerisms during his combine press conference. OK, that last one is not in the model, but c’mon! The after-the-catch production numbers are genuine, but at pick 11 or 12 our prediction is calling this a coin flip, and coin flips at first-round prices have a long track record of not working out.

The Mispriced Asset:

Kenyon Sadiq is a tight end from Oregon and is projected around pick 19. His profile is almost identical to Tate’s, 20.1% boom, 11.9% bust, predicted z of 0.17. His movement in space is elite for his position, as evidenced by his use in the run game. ESPN’s TE projection system, separate methodology, different data, has Sadiq as the highest projection in the system’s history. Two independent models pointing at the same player is worth taking seriously, or could the ghost of Vernon Davis be calling.

Why cones and rushing stats?

The 3-cone measures the ability to change direction in a tight space, and it’s a route running proxy more than a combine parlor trick. Players who post sub-7.0 times move like athletes who can create separation. This might be a little more than just correlation because the player with the best 3-cone drill in recent memory is also now the highest paid WR in the game.

Rushing yards for a wideout or TE aren’t about carries, they’re capturing how well a player operates in space off of run action. A proxy for the kind of physicality that separates producers from depth-chart fillers.

Would you go big on a coin flip?

Tate is good. The model agrees. Lemon and Tyson are coin flips at round-one prices and the model is cooler on both than the market is.

And sitting 13 picks after Tate, at a position that produces more surplus value when it hits, is a tight end with an identical boom/bust profile to the best receiver in the class. The teams paying top-12 for depth are buying exactly what the media sold them. The team that takes Sadiq at 19, or trades up ahead of Lemon and Tyson, is pulling off the Big Short: reading the same data everyone has access to, and reaching a different conclusion. The best receiver in this draft might just line up with his hand in the dirt.

In baseball, the Mendoza Line is the .200 batting average, the bare minimum for a major leaguer to keep his job. Named after Mario Mendoza, it became shorthand for the floor for big league competence.

In the 2026 NFL Draft, there’s a new Mendoza Line. Except this one isn’t the floor. It’s the ceiling. And the draft pundits are pretending a quarterback who didn’t didn’t even approach it should have gone first anyway. We really are in the “silly season” now.

Come again?

Mike Greenberg, on ESPN’s “Get Up,” said it plainly: if Arch Manning were in this draft class, he’d be the number one overall pick over Fernando Mendoza. Dan Orlovsky went further on The Dan Patrick Show, “not even close,” he said. Manning would be the “runaway number one pick. The pick would be made already.”

Not even close?

A Heisman winner. A national champion. The FBS leader in passing touchdowns, passing efficiency, and QBR. A quarterback who went 16-0. And the argument is that a guy who went 10-3 with a 61% completion rate would leapfrog him, and it wouldn’t even be competitive?

Dan is right by magnitude, it’s not close, but he’s definitely wrong by direction. That’s not analysis. That’s nothing but last name recognition and vibes.

The comparison

Here’s what actually happened in the 2025 college football season. Same year, Same sport, Different result.

Mendoza led the FBS in passing touchdowns, passing efficiency, and QBR. He won the Heisman in a landslide, first in all six geographic regions, the first player to do that since 2022. He won the Maxwell, the Davey O’Brien, the Walter Camp, AP Player of the Year, Big Ten Offensive Player of the Year. He led Indiana to its first national championship in school history, going 16-0 and beating Ohio State, Penn State, Alabama, Oregon, and Miami along the way.

Let that sink in again, INDIANA won the the national championship in football!

Manning’s year was solid, but not spectacular. He improved down the stretch. He beat Michigan in the Citrus Bowl. Let’s be honest though, to say that Michigan was a program in turmoil at the time is being nice, so that win may not be as solid of a resume builder as one might think.

The completion percentage gap alone should end this conversation. Mendoza completed 72% of his passes. Manning completed 61.4%. That’s not a difference of opinion. That’s ten percentage points of accuracy, which in the NFL translates directly to drive sustainability, third-down conversions, and wins. Orlovsky knows this. Greenberg knows this. They just don’t care, because they’re not making a data argument. They’re making a vibes argument.

Here’s one more data thing the “Manning goes first” crowd has to explain. The quarterback Arch couldn’t beat out for two years, the highest-rated recruit in 247Sports history, a perfect 1.0000 composite that only Vince Young has matched at the position, went 231st overall last April. Quinn Ewers had 35 career starts, 3,472 passing yards his final season, 31 touchdowns, and led Texas to back-to-back CFP semifinals. The NFL watched all of it and said seventh round. If that’s what the market pays for a Texas quarterback with a top-tier pedigree and questions about consistency, what exactly is the case for Manning going 225 picks higher on fewer starts, a lower completion percentage, and no playoff appearances?

History doesn’t repeat itself, but it sure rhymes

If you’ve been reading Merrittocracy, you know the pattern. Since 2010, quarterbacks with 15 or fewer career Division I starts have a 75% bust rate when drafted in the first round. Trubisky, Haskins, Richardson, the graveyard is full.

Arch Manning has an all too familiar number of starts, 15. Twelve regular season games and a bowl game in 2025, plus two spot starts in 2024 when Quinn Ewers was injured. That’s it. That’s the résumé everyone wants to draft first overall.

The only quarterback to beat the 15-starts curse was Cam Newton, who was 6-6, 250 pounds, rushed for 1,473 yards, won the Heisman, won a national championship, and had a full season of competitive reps at Blinn College before his single year at Auburn. Newton wasn’t an exception to the rule. He was a once-in-a-generation freak who barely survived it.

Manning rushed for 399 yards. Solid dual-threat ability. Not Newton. Not close to Newton. Not in the same area code as Newton.

I love it when a plan comes together…

Here’s the part that should embarrass every analyst pushing this narrative.

In August 2025, before the season even started, Archie Manning told Texas Monthly that Arch wasn’t going to declare for the 2026 draft. “Arch isn’t going to do that. He’ll be at Texas.” Arch pushed back publicly, “I don’t know where he got that from, he texted me to apologize”, and said he was taking it day by day.

Then he did exactly what Archie predicted.

In December, Cooper Manning texted ESPN to confirm: “Arch is playing football at Texas next year.” Arch himself explained the decision: “I felt like I developed a lot this year, especially towards the back half, and I want to keep it going. There’s no reason to leave.”

The Mannings are not a normal football family. They are THE football family. Hello McFly, it’s called the Manning Passing Academy. Between them, they’ve seen every version of this story play out, the early declares who weren’t ready, the guys who went back and became franchise quarterbacks, and the ones who got talked into leaving too soon by a media machine that profits from the hype cycle regardless of what happens next.

Archie Manning didn’t need a boom/bust model to know that 12 starts in the SEC wasn’t enough. He’s lived the data. When the patriarch of the most decorated quarterbacking family in NFL history tells you his grandson needs another year, maybe the correct response isn’t “but the projection though.”

Projection vs. Product

Strip away the TV production and the confident delivery, and the Manning-over-Mendoza argument reduces to three words: talent, projection, bloodline.

Talent is real. Manning has a big arm, he can move, and nobody is disputing the physical tools. But projection is just a polite word for “the numbers don’t support this yet.” And bloodline is not a measurable. Peyton was a four-year starter at Tennessee. Eli was a three-year starter at Ole Miss. Neither of them entered the draft after one season of starting. The Manning family template is patience and development, the exact opposite of what Greenberg and Orlovsky are advocating.

Mendoza, meanwhile, isn’t a projection. He’s a finished product. Thirty-five career starts across Cal and Indiana. A 72% completion rate in a pro-style offense. Forty-one touchdowns against six interceptions. A national championship with the game’s Most Valuable Offensive Player award. An NFL-ready passer who dissects coverages, processes quickly, and wins the biggest games on the biggest stages.

When the argument for one quarterback requires ignoring the stats and the argument for the other quarterback IS the stats, the choice isn’t close.

The Mendoza Line

In baseball, falling below the Mendoza Line means you probably shouldn’t be in the lineup. In the 2026 NFL Draft, the Mendoza Line is what every other quarterback in this class is measured against, and nobody comes close to clearing it.

Fernando Mendoza set the standard. Arch Manning’s family had the good sense to recognize he wasn’t there yet. The only people who can’t see it are the ones paid to talk about it on television the week before the draft.

The Raiders will take Mendoza first on Thursday. And sometime next April, Arch Manning, with another full season, another 12-plus starts, and a developmental arc that looks a lot more like his uncles’, will probably be in the conversation for real.

Patience isn’t weakness. Just ask a Manning.

Data via ESPN, Pro Football Reference, and Sports Reference. Analysis and code at github.com/merrittocratic/nfl-draft-model.

ESPN’s final Rookie of the Year straw poll came in 80-20. Cooper Flagg got the 20.

Let that land for a second. The consensus #1 pick. The biggest preseason ROY favorite in the history of the betting market. The first rookie since Michael Jordan to lead his team in points, rebounds, assists, AND steals. Twenty votes out of a hundred.

Everyone else went Kon Knueppel. I get the case, but when the other guy is mentioned with MJ as the only other “recent” player to do something amazing over a season, the conversation should be over.

The case for Konn

It’s real. Knueppel led the the entire NBA, not just rookies, in three-pointers made with 273. Rookie record. First rookie ever to lead the NBA in triples, period. He shot 43% from deep on volume we haven’t seen from a first-year player since Steph Curry. He played 81 games. He helped drag Charlotte from 19 to 44 wins.

Durable, efficient, winning. The résumé voters reward.

And yet.

It looks close, but is it really?

Stare at the total points row for a minute. Knueppel played ELEVEN more games than Flagg and logged 207 more minutes, and ended up with only twenty-nine more points. One decent scoring night covers his entire volume advantage.

Flagg leads every production category that doesn’t coalesce around the 3-point line or require more games played. He became the youngest player in NBA history to score 50, and almost to prove a point, he did it again four days later. That isn’t a counting stat. That’s an event.

Konn-text matters

The usual pro-Konn defense is contextual. He plays for a real team, Flagg plays for a roster in freefall, team success matters.

Fine. But context cuts both ways. Flagg carried a 26-win Mavericks team that traded Luka last year and then traded his replacement this year. What was billed as an ideal landing spot because Flagg wasn’t supposed to carry the team as a rookie, became anything but.

“Street Clothes” and Kyrie were exactly who we thought they were, and Flagg drew the top defensive assignment every night. Meanwhile, Knueppel was able to lean into his style and play off Ball, literally and figuratively, in an offense that asked him to spot up, move the ball, and take what came to him.

The nearest parallel to this race was Wembanyama vs. Holmgren in 2024. Holmgren played 82 games on a 57-win Thunder team. Wemby played 71 on a 22-win Spurs team. Wemby swept all 99 first-place votes. The per-game alpha on the losing team won unanimously.

Same shape of race. Different answer. What changed?

This ceiling might not be made of glass

Here is the real question, and somehow nobody is asking it: what are these guys going to be next year and in the future?

Knueppel is a 20-year-old who just shot 43% from three, and a league-leading amount of makes in a real NBA offense. Legitimate calling card. Also roughly what he is. Would anyone be surprised if Knueppel settles in as an 16-18 ppg player for the next decade? A very good complementary piece, the third-best guy on a contender, the kind of sharpshooter who plays fifteen years and makes real money doing it. Great career. Not a franchise-tilter.

Flagg is a 19-year-old who just put up 21/7/5 on a team with no floor spacing and relied on him to be the primary creator. Put shooting around him. Drop in a secondary offensive engine so defenses can’t load up on him every possession, and a leap to 25-27 ppg isn’t optimistic, it’s the expected outcome. The ceiling on a 6-9 forward who defends, passes, and can drop 51 against a playoff-caliber defense isn’t the ceiling of a player, it’s the ceiling of the league itself.

One guy is likely approaching his ceiling, while the other is just starting to find his.

Who do you want with the ball?

One more question, and it’s the one Rookie of the Year should measure. It’s late, it’s tied, you need a bucket. Flagg or Knueppel, who do you want?

Nobody’s honest answer is Knueppel. Not the voters, not the gamblers, and given how the play-in went, not Charles Lee.

Tuesday in the play-in against Miami, Knueppel went 2-for-12 from the field, 0-for-6 from three, and watched crunch time from the bench while LaMelo dragged them to an overtime win. Friday in Orlando, the same team Flagg dropped 51 on, 3-for-10, 1-for-6 from deep, 11 points in a 121-90 elimination that ended Charlotte’s season and extended the longest active playoff drought in the NBA.

This should concern the voters who went 80-20 for Konn, because it’s rare to see a consensus take go this bad, this fast. The alpha of this year’s class was Flagg. The per-game numbers say so. The degree-of-difficulty says so. The ceiling says so. And when the lights came on, the bench said so.

Don’t be Konned.

Data via NBA.com, ESPN, and Basketball-Reference.

Weeks upon weeks of discussions about pure pass rushers vs. hybrid edge players vs. traditional off-ball linebackers, the draft industrial complex loves to exhaust a story line. The problem is, some of these decisions really do come down to coin flip odds, or worse. Given the discourse, you would not be crazy to think there is nothing but sure things to draft along the front seven.

And this is the tell.

Ask yourself, when is the last time you’ve heard any of the analysts have an in-depth discussion on interior defensive linemen?

The position that “has a really nice personality”

While this position would never be considered sexy, unless you’re Dexter Lawrence, interior defensive line is the safest premium bet in the front seven, and it’s really not close. If you recall the article on Downs’ value position, the boom / bust differential was effectively a push for Defensive Linemen. The data magic happens when you separate EDGE from IDL in that analysis, and you quickly see a pattern emerge.

First-round IDL posts a +10.9% boom / bust differential and that is, by a wide margin, the best outcome profile of any position in the front seven. It’s also the only front-seven position that starts out and remains positive across all rounds (+8.1%). This isn’t a round-one fluke or a small-sample artifact. It’s a structural feature of how the position translates from college to the NFL.

The boom list reads like an All-Pro roster, with most bound for Canton. Aaron Donald at 13. Ndamukong Suh at 2. Haloti Ngata at 12. Fletcher Cox at 12. Jeffery Simmons at 19. While Chris Jones slipped to the second round, he deserves honorable mention here, because he became one of the best defensive tackle of his generation. When you work through the names, the pattern is: if you take a good college interior defensive lineman in the first round, you get a good NFL interior defensive lineman. The position rewards the investment.

The bust list is where it gets interesting. Dontari Poe, Jarran Reed, Robert Nkemdiche, Malik McDowell, Justin Harrell. Go read the stories. The pattern is injuries and off-field issues, not talent misses. When an IDL busts, it usually means something went wrong around the talent. That’s a fundamentally different risk category than an evaluation miss like “that quarterback just can’t see the field,” and the problem is the market doesn’t price the distinction.

Unless you already have two monsters in the middle, most teams drafting from 15 - 32 would be better off drafting one of the elite interior defensive lineman in this draft. The team who does is getting the position with the best risk-adjusted return in the front seven, at a price the market has been undervaluing for a decade and a half.

Nobody will write a think piece about the pick, and that’s part of the reason it works.

The position that drives the GMs wild

Despite the flash, when you flip to edge rusher, the picture inverts.

First-round EDGE posts a −11.3% boom / bust differential that is, by the same margin, the worst outcome profile in the front seven. When you draft one in the top-three picks, it can get Coyote Ugly real quick. Dion Jordan. Dante Fowler. Vernon Gholston. Solomon Thomas. Clelin Ferrell. Top-shelf investment, bottom-shelf returns, and decade-spanning consistency across different teams and different front offices all making the same mistake at the same position with the same confidence.

But here’s the catch that reframes everything: all-rounds EDGE produces a +0.8% differential. Essentially break-even. The position isn’t broken. The premium is.

When you look at a scatterplot of first-round EDGE production, the story gets sharper.

The intuition would be that booms cluster in one region and busts in another, that you could look at college production and see which prospects were the safe bets. That’s not what happens. The blue dot in the upper-right corner is JJ Watt. Eleventh overall pick, 2011. Elite sack production, elite tackle production, the best defensive player of a generation, and his z-score of 5.53, is so far off the charts it breaks the scale. He’s the ceiling of what the upper-right quadrant can produce. And the red dots sitting right next to him, same quadrant, same elite production profile, are first-round busts. The ceiling is real. The guarantee isn’t.

Then look at the lower-left. Robert Quinn is sitting at near-zero on both axes, and he boomed. His college stats aren’t a talent story, they’re a suspension. Quinn missed his final college season, so the production numbers read as empty, but the talent was there before the year off. The chart says “non-producer who hit.” The reality is “elite talent whose stats got zeroed out by circumstances.” Context matters.

Now look at where this year’s prospects land. Bailey Reese, and Bain.

David Bailey is the most instructive case, better than 90th-percentile sack production but below-median total tackles. A pure sack specialist. And the red bust dot right next to him? That’s Chase Young. Second overall pick, 2020. Same profile: dominant sack numbers, relatively modest broader production. Young was supposed to be the generational pass rusher, and he never became the player that pick demanded. Bailey at the top of the draft is the same bet the league already made on Young, in the same neighborhood of the scatter, at the same price point.

Reese is the opposite profile: elite on overall production and far better than median on sacks. He’s parked in the upper-right where the ceiling is JJ Watt and the floor is the bust dots sitting next to him. Bain lands between Bailey and Reese, solid but not elite on either dimension, the kind of prospect that the data says can go either way.

None of those positions on the chart tell you who hits. That’s the point. The graphic isn’t a map, but it is evidence of the position’s volatility in the first round, regardless of production profile. Teams aren’t picking the wrong prospects. They’re paying premium prices for a position where the evaluation tools can’t reliably separate hits from misses, because the best defensive player of a generation and first-round busts occupy the same space.

This is what makes the Day 2 value curve the conclusion, not the footnote. Frank Clark at 63. Calais Campbell at 50. LaMarr Woodley at 46. These aren’t outliers, again they’re evidence. Round-one and all-rounds EDGE produce different numbers because Day 1 and Day 2 are drawing from the same distribution of outcomes, and only one is priced like a premium investment. The position is fine. Paying top-five for it is where front offices keep walking into the same wall, not because the scouts are bad, but because the price is wrong, and has been wrong for fifteen years.

Just when you thought it was safe

If round-one EDGE is the bet everyone knows is a coin flip and takes anyway, linebacker is the bet everyone thinks will work out and doesn’t.

First-round LB posts a −9.3% differential. Second to only edge rushers for the worst differential in the front seven. The bust list is visceral in a way the EDGE list isn’t, because the picks were even more premium: Aaron Curry at 4. Barkevious Mingo at 6. Rolando McClain at 8. Isaiah Simmons at 8. Top-ten capital, never became the player the consensus promised.

And here’s the counter-intuitive part: all-rounds LB is +3.9%. Positive. The position is fine at price. The round-one tax is where the carnage lives.

This is the rare narrative-check where the data agrees with the conventional wisdom. “Don’t take linebackers in round one” is a thing draft analytics types have been saying for years. The data backs it up cleanly, and the more interesting question is why the draft-industrial complex keeps making the pick anyway. The answer is usually some variation of “but this guy is different”, and every year, different is what produces −9.3%.

The chess piece archetype is the warning on top of the warning.

If round-one LB is already a coin flip with the house edge, asking a first-round linebacker to be a positional chess piece is deployment risk layered on top of positional risk. Isaiah Simmons at 8 in 2020 is the cautionary tale, and he wasn’t a talent miss, he was a deployment miss. A player asked to be everything who never got to be one specific thing long enough to become great at it.

The players who turned the chess-piece label into production all did the same thing: they ignored the label. Micah Parsons went 12th in 2021 as a “versatile LB,” then Dallas handed him an edge role and the rest is history. Haason Reddick was a bust as a linebacker before he reinvented himself as a pure pass rusher and became one of the best in the league. Khalil Mack was drafted as an outside linebacker, became the first player in NFL history named First-Team All-Pro at two different positions, and did it by leaning into the edge role, not away from it.

The pattern is consistent: the chess-piece label is a warning, and the players who beat it did so by eventually committing to a position. The ones who stayed “versatile” forever are in the bust column.

Three bets, one right answer

Bet One: The Chess Piece. Sonny Styles and Arvell Reese are the headliners. The athletic profiles are legitimately jaw-dropping, Styles’ 43.5-inch vertical and 4.46 at 244 lbs, Reese’s matching 4.46 at 6’4” with a 3-cone drill that is better than most WRs. The pre-draft machine is running the same play on both of them it ran on Isaiah Simmons in 2020: they’re versatile, they’re weapons, they’re chess pieces. Admittedly, there is ambiguity on Reese’s draft position. He’s included here because he’s listed both as a LB and EDGE depending on the draft site. Danger, Will Robinson!

Here’s what the model is actually saying about Styles. The prediction isn’t anchored on the vertical or the 40. It’s anchored on being a highly productive college defender. The combine is the sizzle. The college production is the steak. And underneath both, Styles is still a first-round linebacker that could be asked to be a chess piece, two warnings, stacked. The team that drafts him has to commit him to a role early. The Dallas-and-Parsons playbook, not the Arizona-and-Simmons playbook.

Bet Two: The Pass Rusher. Rueben Bain and David Bailey are the cleaner value propositions on conventional paper, players who will line up on the edge and rush the quarterback, no position ambiguity required. But the round-one EDGE numbers don’t care about clean profiles. Almost everyone on the first-round EDGE board has a clean profile. That’s the floor, not the ceiling. If Bailey goes in the top five, the model’s position on him is the same position the data has taken on top-five edge rushers for a decade: this is the exact pick that is littered with busts, and nothing in the evaluation profile can tell you whether he’s the one who breaks the pattern.

A little more on this. The Jets hold the second pick. The consensus mock has Arvell Reese. The analytics case for Bailey instead isn’t just about positional archetypes, it’s organizational. New York’s team development data over the last decade grades out at as slightly better than average for defensive linemen across 6 picks. It’s a quiet, but real signal. They know what to do with a player who lines up and rushes the passer. Their linebacker number, however, is WELL below average across 2 picks. Small sample. Unmistakable direction. The Isaiah Simmons warning in the previous section isn’t hypothetical for this franchise, it’s the template. A player asked to be everything, by a coaching staff that couldn’t commit him to one thing, who never got to be great at anything specific. Bailey is the pick that plays to what the Jets’ track record actually shows. Reese is the pick that asks the Jets to be a different organization than the data says they are.

Bet Three: Just Play Linebacker. CJ Allen and Jacob Rodriguez are the least exciting players in this conversation, and under a different set of numbers they’d be the boring safe pick, sideline-to-sideline, three-down capable, NFL-ready scheme fits. Under the real numbers, they’re the boring pick at a position that busts 33% of the time in round one. Allen leading Georgia in tackles and winning Butkus consideration doesn’t make the base rate go away. Rodriguez’s seven forced fumbles don’t make the base rate go away. The position taxes premium capital regardless of archetype.

This doesn’t mean don’t draft them. It means the “just play linebacker” framing isn’t a safety valve. It’s a different flavor of the same first-round tax.

The right answer is none of the above. The right answer is the interior defensive lineman the same team could take five picks later, at a position the market has been mis-pricing for fifteen years, while everyone argues about whose chess piece is the real chess piece.

The bottom line

The front seven in this draft is loaded with talent and priced exactly the way every front seven gets priced: EDGE at the top, linebackers wherever the best-player-available board lands them, interior defensive line somewhere in the middle of the conversation and nowhere in the headlines.

The data says flip it.

The interior defensive lineman in the back half of round one is the safest premium bet on the board. Round-one EDGE and round-one LB are both coin flips with the house edge, for different structural reasons, EDGE because the evaluation tools can’t separate the filtered population, LB because the position taxes premium capital regardless of which archetype you bet on. The team that walks away from this draft with the best front seven probably isn’t the one that made the loudest pick. It’s the one that took the interior defensive lineman and went to dinner.

Whichever team does spend first-round capital on Styles, Reese, Bain, Allen, or Bailey, the outcome won’t just be about the player. It’ll be about whether that organization has a track record of committing first-round defensive talent to a defined role. The front-seven picks will mean different things depending on whose name gets called first and where they go.

Draft night is April 23. The Model is Locked and Loaded. Data via Pro Football Reference. Analysis and code at github.com/merrittocratic/nfl-draft-model.

The regular season is a wrap and the play-in games begin soon. I think someone will win a postseason award, but at this point you can’t be too sure. The NBA playoffs officially start April 18th, and most NBA narrative creators have already made up their mind about Victor Wembanyama. He won’t win the championship this year. The Spurs are too young, the West is too deep, the sample size on playoff Wemby is the same as the sum of Bluto’s and Flounder’s grade point averages, 0.0. The consensus isn’t necessarily wrong, but the way people are talking about the narrative is almost more interesting than whether it’s true.

Because underneath the “he’s not ready yet” framing is a question nobody is actually asking out loud: has a player like this ever done this before? We’re not just talking about a young player, or a good player making a surprising run. This is a generational big man, in year three, carrying a 60-plus-win team into the playoffs for the first time. Has this specific thing happened, and what happened when it did?

There is one real answer. It has two chapters. And, ZERO mentions.

Chapter One: Milwaukee

In 1969, the Milwaukee Bucks used the first overall pick on a 22-year-old center out of UCLA named Lew Alcindor. He won Rookie of the Year. The Bucks made the playoffs and lost. Not what I’m referencing. Now, in year two, they went out and got Alcindor a co-star: a 31-year-old future Hall of Famer, one of the greatest point guards in NBA history, who was the first player to average a triple-double in a season, Oscar Robinson.

Domination ensued, as Milwaukee went 66-16, ran off a then-record 20-game winning streak, and swept the Baltimore Bullets 4-0 in the Finals. Alcindor, who would take the name Kareem Abdul-Jabbar the following offseason, won the regular-season MVP, the scoring title, and the Finals MVP.

I really hope this isn’t a glitch in the Matrix, because these numbers look really similar:

The wins are comparable. The age is comparable. The accolades are comparable. But, there’s always something isn’t there, and that something in this case is the last row.

De’Aaron Fox is a good player. He’s averaging 19 points and 6.3 assists per game this season, made his second All-Star team as alternate, and has been exactly what the Spurs needed when they traded for him. Oscar Robertson averaged 25.7 points, 9.5 rebounds, and 7.4 assists per game over his stellar career, and while his performance dipped in this championship run, he is widely considered the most complete player in the history of the position. The Big “O” and De’Aron are definitely not the same player, and the distance between them is most of what separates “Wemby could win this” from “Wemby wins this.”

Kareem needed the right co-star to make it happen, and Oscar fit the bill.

Chapter Two: Los Angeles

Here’s where the story comes full circle, because it doesn’t just end in Milwaukee.

Kareem was traded to the Lakers in 1975. He put up enormous numbers, won MVPs, but the Lakers went nowhere in the playoffs. Four years of transcendent individual performance and zero championships, because they could never find the right co-star.

Then, in 1979, the Lakers drafted a 6’9” Spartan point guard named Magic Johnson with the first overall pick. In Magic’s rookie year, his first playoff run, the Lakers met the Philadelphia 76ers in the Finals. Kareem’s ankle gave out in Game 5. Magic, just a 20-year-old rookie, started at center in Game 6, played all five positions, and dropped 42 points, 15 rebounds, and 7 assists to close out the series. He won the Finals MVP. Kareem, becoming more of an elder statesman at 33 years-old, finally had his second ring.

Two co-stars. Separated by almost a decade. Both Hall of Famers, both in the conversation for greatest ever at their position. It took both to deliver titles to the greatest big man of his era, who some consider to be on the Mt. Rushmore of basketball.

What It Would Take

Wemby winning a championship this year would be more historically unprecedented than Kareem’s run in Milwaukee, he’d be doing it in his first playoff appearance, a year later in his career, with a younger supporting cast against a deeper conference. The “it’s never been done quite like this” framing isn’t hype. It’s accurate.

A deep run in this postseason might be the most honest ceiling for Wemby and his Spurs. A Conference Finals appearance would already be historically elite for a player in this situation. And a championship, while not impossible, would require the “Alien” to do something no dominant young big man has done without a co-star on their way to the Hall of Fame.

Fox is good. Fox is not Oscar, and Fox is definitely not Magic.

That’s not a knock on Fox, or on this Spurs team. It’s just the only honest answer to the only real precedent.

The most interesting version of this story isn’t what happens in April. It’s whether the Spurs find their Oscar, or their Magic, and what Wemby looks like when they do.

We’ll be running NBA playoff narrative-checks through June. Bad conventional wisdom doesn’t take the summer off.

Rory. Jordan. Scottie. The name changes, but the answer never will.

There isn’t one. There won’t be one. And once you understand why Tiger was Tiger, you’ll stop expecting one.

Tiger Woods wasn’t just a talented kid who got really good at golf. He was a project — conceived, designed, and executed with a level of intentionality that borders on being unsettling. What happens when you have a Dad who was both trained in psychological warfare and taught how to be media savvy. The answer, you now have someone understands both how to break a mind and how to craft a media narrative around it. The real, real answer is, you get Earl Woods, and he applied both those skill sets to his son starting in the highchair.

The psychological conditioning is well-documented — the trash talk mid-backswing, the deliberate distraction, the pressure manufacturing — all designed to build a competitor who couldn’t be rattled on the biggest stages. Earl didn’t just teach Tiger golf. He built Tiger’s brain for golf. Add to this equation Kultida Woods — Thai-born, deeply protective, fiercely private — and now you have the perfect home environment to keep the project sealed. The combination gave Tiger an almost hermetic early life: channeled, shielded, optimized.

You don’t accidentally produce that. You could barely produce it on purpose.

Now, a look at most people’s next Tiger, Rory.

As of this writing, he is on the back nine at Augusta on Sunday, fighting for a second consecutive green jacket after doing something Tiger never did: losing the largest 36-hole lead in Masters history on a Saturday and then showing up anyway on Sunday.

Rory’s dad, Gerry McIlroy, was a regular guy. Bartender. Factory worker. By most accounts, he worked multiple jobs and reportedly took out loans to fund his son’s golf development. No psychological warfare certification. No media training. Just a Dad who believed in his kid and figured out how to help make his dream become a reality.

The result? A golfer who is, by every public account, a functional human being with a healthy marriage, a daughter he adores, and a genuine relationship with the sport — not just a weaponized version of it. Yes, things have not been perfect, but to my knowledge, Rory has never been threatened with a 9-iron.

Before last year, Rory spent 14 years carrying the weight of failure before finally winning the Masters in a playoff. It was not because he was engineered to never feel pressure. Instead, it was because he learned, the hard way, to carry it like a human being.

What happened Saturday — a six-shot lead, evaporated — would have looked familiar to anyone watching what happened to him 2011. But Rory came back on Sunday. Battling. Still in it. That’s not a Tiger story. That’s a BETTER story, in a lot of ways.

The formula that made Tiger elite is also the formula that made his personal unraveling so spectacular. When you build a person for singular, obsessive focus — when the psychological architecture is all ballistic trajectory and no lateral load tolerance — the highs are generational and the lows are catastrophic.

Tiger gave us both. Fully. No refunds.

The analytics lesson here is one Merrittocracy returns to constantly: when a data point looks like an outlier, the right question isn’t “who’s next?” It’s “why was this one so extreme?” Once you understand the inputs, you stop looking for replication and start appreciating the singularity.

Rory might win today. He might lose it again. Either way, he’ll go home to his family and come back next year. That’s not a consolation prize. That’s what a healthy outcome looks like.

We got one Tiger. We got every chaotic, brilliant, broken, transcendent thing that came with him.

That’s going to have to be enough. And honestly? It is.

I’m talking about Running Backs… What did you think I meant? Last week we talked about how the league was about to undervalue Caleb Downs because Safety is not just the position, it’s the mindset of too many NFL front offices. This week, different position, same disease.

Jeremiyah Love is going to go in the top 10 of the 2026 NFL Draft. Maybe, even inside of the top 5. Todd McShay recently told Joel Klatt that Love is the first running back he’s ever had ranked #1 on his big board in 26 years of evaluating. Awesome news for Love! But then, in the same breath, he also said that he’s “not fully convinced” about Love at 5 to the Giants. So, “the best player on the board”, and the guy who put him there still isn’t sure it’s OK to put him that high.

Here’s the thing, Love should go in the top 5. Our model agrees, the data agrees, and history agrees.

The problem isn’t the pick. The problem is the language. To justify taking a Running Back in the top 5, the draft-industrial complex has had to invent a workaround. Love isn’t a running back — he’s an “offensive weapon.” A “joker.” A “matchup nightmare.” A “swiss army knife.”

He’s a Running Back. And it’s fine to call him that. It’s been fine for a long time. We just have to start admitting it.

The laundering

In Goodfellas, Paulie didn’t want his name on the restaurant. He just wanted the money. So they ran everything through a front until nobody could trace it back to what it actually was. That’s what analysts do with the Running Back position label. The pick is a Running Back, but by the time it goes through the mock draft machine, it comes out the other side as everything but.

Let me show you what I mean. Pull up any top-10 mock from the last month and read the Love comments out loud. You’ll find some version of these phrases:

• “He’s more than just a running back.”

• “He’s a weapon you can move all over the formation.”

• “He’s an offensive chess piece.”

• “He’s not in the same conversation as other backs because of what he does in the passing game.”

Here’s what those sentences have in common: every single one is a defense against an argument that nobody is actually making. Nobody thinks Love is bad. Nobody thinks he’s a low-ceiling player. The thing being argued against is the position label itself.

We’ve quoted McShay, so we might as well quote his older, alter ego. “He’s so much more than just a running back” — Mel Kiper, in ESPN’s collaborative mock, literally said: “I know, I know. Kiper doesn’t like early-pick Running Backs. But Love is so much more than just a Running Back.”

You only need to launder a label when the label is the problem. And the label is only a problem because the analytics community spent the better part of a decade building a case that Running Backs aren’t worth premium picks, a case the public draft discourse internalized so completely that even the analysts who love Love feel obligated to pretend he’s something else.

By the way, I really want to hear Jeremiyah go full TO and say, “I love me some Love…”

The laundering tells you everything. If Running Backs were genuinely fine to draft in the top 5, everybody would be OK with using the R word again.

Where running backs actually live

First-round Running Backs boom at roughly the same rate as first-round receivers, both sit in the low-to-mid 20s. This is the position the orthodoxy says you should never touch in round one. The data doesn’t agree.

But there’s a second number that matters even more, and it’s the one that breaks the “draft a wide receiver instead” argument in half.

Look at the y-axis (vertical axis). When first-round Running Backs hit, they hit bigger than first-round receivers — roughly 29 average 4-year Approximate Value (AV) versus 22. That’s not a rounding error. That’s an entire tier of production the “draft a receiver instead” crowd is leaving on the table. A great Running Back can carry an offense in a way that a great receiver, no matter how dominant, structurally cannot.

Nearly the same boom rate. Seven more points of AV when it hits. And a position the market is currently discounting relative to what the data says it should be worth.

That’s not “non-premium.” That’s a market inefficiency hiding behind a vocabulary problem.

The bust list everyone keeps citing

The case against drafting Running Backs in the first round is built almost entirely on a small group of names that get repeated until they sound like a category. Trent Richardson. Rashaad Penny. Najee Harris. Sometimes Ronald Jones. Sometimes Mark Ingram, depending on how aggressive the argument is being applied.

Let’s actually look at that list.

Trent Richardson went 3rd overall in 2012 and washed out by year three. He is the patron saint of the “never take a Running Back high” movement, and his bust was so spectacular that it hardened a generation of analytics writers. Real bust. Counts.

Rashaad Penny went 27th overall in 2018 and never produced. Real bust, also counts — but he was the 27th pick, not a top-10 swing. The cost of being wrong was a late-first, not a franchise-defining whiff.

Najee Harris went 24th overall in 2021 and has been… fine. Not a star, not a disaster, exactly the kind of Pittsburgh workhorse back the Steelers thought they were drafting. Consistently producing 1,300-plus scrimmage yards and 8 total TDs per year since being drafted is not ELITE, but it’s definitely not a bust. Coming out of Alabama, he actually had the combo of receiving and rushing yards the NFL is looking for in a chess piece. The “Najee was a bust” line was wrong before he was drafted and it’s still wrong now, despite his lost season with the Chargers.

Now look at the same window for receivers. Henry Ruggs, John Ross, Kevin White, Corey Davis, Laquon Treadwell, N’Keal Harry. These are all top-15 wide receivers from the last decade who didn’t work out. Some of them spectacularly. The wide receiver bust list is longer than the Running Back bust list, and the busts cost more in terms of expected production lost. The difference is that nobody built a movement around them, because the position carries social permission to be drafted high.

The Running Back bust narrative is built on three or four memorable names. The wide receiver hit narrative is built on Justin Jefferson and Ja’Marr Chase. Both narratives are doing a lot of cherry-picking, and the cherry-picking happens to align perfectly with what the analytics community already wanted to believe.

And yes, it was very hard to not list ALOT of Ohio State Wide Receivers there as hits, but I opted for a little more objectivity.

What about CMC, Gibbs, and Bijan?

If we’re going to be honest about the busts, we have to be honest about the hits, and the hits are louder than anyone wants to admit.

Christian McCaffrey went 8th overall in 2017. The pick drew positional-value howling from the day it was made. McCaffrey has been the most versatile offensive player in football for over half a decade, won Offensive Player of the Year, and is the kind of player teams build entire offenses around. His 2023 season in San Francisco was the closest thing the modern NFL has produced to a one-man offense.

Bijan Robinson went 8th overall in 2023. The Falcons were mocked relentlessly for the pick. Two years in, Robinson is averaging over 1,400 yards from scrimmage and is one of the few legitimately good things on a roster that has spent two seasons searching for an identity. Take the pick away and Atlanta is almost unwatchable. I mean, they’re bringing in Tua because he’s left-handed. Desperate times….

Jahmyr Gibbs went 12th overall in the same 2023 draft, and this is the comp that matters most for Love. The Lions caught even more grief than the Falcons did, because they had needs at premium positions and used a top-12 pick on a running back. Two years later, Gibbs is a Pro Bowler, a Super Bowl contender’s leading offensive weapon, and exactly the dual-threat archetype that the “weapon” framing is trying to describe when it gets clumsy. He runs hard, he catches everything, he scores from anywhere on the field, and Detroit’s offense runs through him in a way that would not be possible if Brad Holmes had listened to the analytics community and traded back.

These aren’t “exceptions that prove the rule.” They’re the modal outcome when a team picks an elite running back near the top of round one and the player is actually elite. The rule is the exception, and the exceptions are the rule. We just keep telling the story backwards because the story sounds smarter when it’s contrarian.

And here’s the part that makes the Love case specifically: every single one of those backs — CMC, Bijan, Gibbs — was also called a “weapon” coming out. The same laundering that’s happening to Love now, happened to all three of them in their draft cycles. The vocabulary trick isn’t new. It’s just the only way the discourse can let itself do something it secretly wants to do.

And the most recent test case is already on the board. Ashton Jeanty went 6th overall to the Raiders in 2025 and immediately put up 1,321 scrimmage yards and 10 total touchdowns as a rookie behind one of the worst offensive lines in football.

But, if you want to anchor on a single archetype for what Love is being drafted to become, look at Gibbs. Same dual-threat profile, same “can line up anywhere” framing, same skepticism from the same people, and the results will be the same two years from now.

The prospect

Let me actually talk about Love for a minute, because the piece so far has been about everyone else’s arguments and not about him.

Love is 6-0, 215 pounds, ran 4.36 at the combine, and spent three seasons at Notre Dame doing things that don’t fit on a stat sheet. He averaged over 7 yards per carry as the lead back in a pro-style offense playing one of the tougher schedules in college football, that included a National Championship run. He caught passes out of the backfield, lined up in the slot, and ran routes that wouldn’t look out of place on a wide receiver’s resume. He’s the rare back who genuinely can do all of the things the “weapon” framing implies — but, and this is the critical part, the things he does best are still Running Back things.

He runs through arm tackles. He has a second gear in the open field. He has the patience to let a hole develop and the burst to hit it before it closes. Those are the traits that make Running Backs valuable in the NFL, and they’re the traits that translate. The receiving usage is a bonus on top, not the main course.

The framing inversion here matters. Love isn’t valuable because he’s a “weapon who happens to play Running Back.” He’s valuable because he’s an elite Running Back who also happens to be a useful receiver. Get the order right and the pick stops needing a vocabulary defense.

The discount that isn’t a discount

Here’s where the Downs piece and this one connect, and why they were always going to be a pair.

Caleb Downs is going to slide because the word “safety” carries a positional-value tax that the data doesn’t justify. Jeremiyah Love is going to get drafted in the top 10 despite the words “Running Back” carrying a similar tax — and the only reason it works is because the discourse has built a linguistic escape hatch that lets the people taking him pretend they’re taking something else.

Both situations are the same underlying problem. Position labels are doing more work in NFL draft rooms than the actual data supports. With Downs, the label is dragging him down past where his talent says he should go. With Love, the label is being scrubbed off his jersey so the people who want to take him don’t have to defend the position.

The honest move in both cases is the same: ignore the label, look at the player, look at the data, and act accordingly. Downs is a top-10 talent at a position with a perfectly defensible bust rate. Love is a top-10 talent at a position with a better boom rate than half the so-called premium positions. Neither of these is a hard call once you turn off the vocabulary filter.

So…, Draft a Running Back then?

We’re not saying that you should go rip up your mock drafts, and put four Running Backs in the first round. There is no Ty Simpson comp in the Running Back group. No Bailey vs. Reese debates. Love is a one-of-one in this draft. The structural value of the Running Back position matters and the analytics community was correct to push back on the era when teams were taking a Running Back in the top 10 every other year regardless of the prospect.

What we are saying is that the correction overshot. The “never take a Running Back in round one” rule isn’t a rule, it’s a heuristic that worked as a corrective when teams were being undisciplined, and has now hardened into orthodoxy that requires some mental gymnastics to override, even when the player obviously deserves it. When the people who are right about the pick can’t say why they’re right without inventing a new word, the framework has stopped helping and started getting in the way.

Love is going to be drafted high. He’s going to produce. The team that takes him is going to look smart. And the post-draft analysis is going to credit the pick to his “versatility” and his “matchup value” and his “ability to line up anywhere,” when the actual reason it worked is that he’s a really good Running Back and good Running Backs are still worth drafting.

Just say the words. He’s a Running Back, and then draft him anyway.

Next up: the front seven. The data on edge rushers and linebackers has some surprises in it, and Sonny Styles is the test case for whether the league has learned anything from how it’s been handling positions like safety.

Data via Pro Football Reference. Analysis and code at github.com/merrittocratic/nfl-draft-model.

Every so often when building a model, you come across a feature that you can’t ignore. The training and retaining of top talent is so important to any business, and the NFL is no exception. The success of Sam Darnold with the Vikings and Seahawks isn’t an accident, and one of the reasons why is hiding in plain sight. Most public draft models ignore the drafting team entirely once the pick is made. That’s a mistake. Some organizations are just genuinely better at turning draft capital into production, and 15 years of data makes that signal visible.

Who’s winning the talent game

Oddly, two franchises re-located to Los Angeles and saw their talent development data immediately improve. A word of caution though, in case you’re thinking you should go ahead and pack up all your belongings, the Raiders are still bad, so maybe the moving van isn’t the answer.

The San Diego “Super Chargers” ranked 27th, a negative residual mean, and a 14.1% bust rate over 71 picks. Talented players, consistently underperforming their draft slots. On the other hand, the Los Angeles Chargers ranked 1st. Best residual mean in the league, with just a 3.8% bust rate. Yes, 27 picks makes for a small sample size, but the direction is hard to dismiss. The organizational reset that came with the move shows up in the numbers.

The Rams tell a similar story. St. Louis finished dead last — 34th, –0.229 residual mean, with a 15.4% bust rate over 88 picks. The Jeff Fisher years were exactly as bad as the record suggested. Even with their supposed de-valuing of the draft, the Los Angeles version of the Rams rank 14th. Les Snead, freed from that organizational inertia has built something that trends toward actually developing the talent that it drafts, despite the narrative.

And, before you say, “Really, the A’ints at number 2?”, remember this covers the Payton/Brees era. Have you seen what Payton is now doing with the Broncos, despite still being under the deep Russell Wilson contract waters? You can argue about the person, but you can’t argue with his results.

Las Vegas appears to be running the moving van control experiment. Oakland ranked 31st — –0.155 residual mean, 17% bust rate. Even with the move to Las Vegas in 2020, the Raiders have rotated through three head coaches, two general managers, and multiple rounds of roster chaos.

Same organization, different zip code, same results.

The LAC and LAR relocations worked because the organizations changed. The Raiders relocated and kept being the Raiders. The moving van isn’t the variable. The front office is.

More struggles hiding in plain sight

New England ranks 29th, and that number is going to surprise people.

The Patriots Development Machine is one of the most durable narratives in football. Here’s what the narrative misses: Belichick’s genius was never in the first round. Tom Brady was a 6th-round pick. Gronkowski was a 2nd-rounder. Julian Edelman a 7th. Malcolm Butler and J.C. Jackson were both undrafted. Matthew Slater, eight Pro Bowls, was a 5th-round pick on special teams. The dynasty was built almost entirely outside premium draft capital.

On offense, fifteen-plus years of first-round picks produced Nate Solder (one Pro Bowl, 2017) and Mac Jones (one alternate appearance in 2022, when Mahomes and Lamar both opted out). The defensive first-rounders held up: McCourty, Hightower, Chandler Jones all delivered. But the overall first-round residual was negative for most of the Belichick era.

That’s not a failure. It’s a system, one that required Brady to work, and worked brilliantly for twenty years. When Brady left, the organizational approach that ranked 29th in draft-pick-adjusted development was suddenly visible. The last three years looked exactly like what the data had been quietly describing all along.

What it means going into Draft Night

Before draft night, Merrittocracy will be publishing team-level development grades alongside our prospect scores. You’ll know which teams picking in the top 15 have the track record to back up their investment, and which ones are about to repeat the same organizational mistakes. The pattern for flagging repeat offenders in the draft has been in the data for a decade.

The drafting team isn’t noise. It’s signal. And it’s been hiding in plain sight.

Data via Pro Football Reference. Analysis and code at github.com/merrittocratic/nfl-draft-model.

The draft has a positional value problem, and it’s about to consume one of the best prospects in this year’s class.

Three of the most universally praised players available this April play positions the NFL has collectively decided aren’t worth premium draft capital: a running back (Jeremiyah Love), a linebacker (Sonny Styles), and a safety (Caleb Downs). What makes this year unusual is that the draft analysts have spent the last month bending their own rules to talk Love and Styles into the top five anyway. Love isn’t a running back, he’s a “weapon.” Styles isn’t a linebacker, he’s a “defensive chess piece.” Did you see their 40 times? Did you see Styles’ broad jump distance?

Downs, on the other hand? He’s just a safety.

Same tier of prospect. Same “non-premium” label. Different treatment. We ran the numbers on 15 years of first-round picks, and the results tell a story the draft-industrial complex doesn’t want to hear.

The consensus

Let’s be fair to the argument before we take a crack at it.

Safety is one of the least-valued positions on draft night. Over the last decade, only six safeties have been selected in the first round. Zero have gone in the top five. The highest was Kyle Hamilton at 14th overall in 2022, and that was considered a mild slide at the time — most evaluators had him in the top 10 on pure talent.

Daniel Jeremiah (DJ) — who has Downs as his top safety and a top-10 overall talent — said it plainly on Dallas radio last week: “Weird stuff happens in the safety position.” He pointed to Downs being under six feet with short arms as reasons teams might pass. The subtext was clear: the position depresses the pick, regardless of the player.

This isn’t just media chatter. It’s reflected in how NFL front offices actually spend. The analytics community has built surplus value curves showing that safety produces one of the weakest returns on first-round investment. The argument goes: you can find comparable talent at the position in round two or later, so spending a top-15 pick on a safety is an inefficient allocation of draft capital. Use that pick on an edge rusher or an offensive tackle, where the surplus value curve is steepest.

It’s a clean argument. It’s intellectually coherent. And it has a blind spot the size of Kyle Hamilton’s contract.

How we measured this

Before we get to the numbers, a quick explanation of the ruler, because the way you measure “good” turns out to matter a lot.

We used Approximate Value (AV), a stat created by Pro Football Reference that puts a single number on a player’s season, regardless of position. Think of it like a universal currency for football production. A quarterback who throws for 4,000 yards and 30 touchdowns earns a high AV. A safety who racks up 100 tackles, three interceptions, and plays every snap also earns a high AV. It’s not perfect — no single number can capture everything a football player does — but it’s the best tool available for comparing across positions, and it’s the standard in draft analytics research.

Here’s the important part. We’re not measuring raw production. We’re measuring whether a player outperformed or underperformed what his position historically produces at that draft slot. A safety drafted 14th isn’t being compared to a quarterback drafted 14th — he’s being compared to other safeties drafted in the same range. That way a safety doesn’t get penalized for accumulating AV differently than a running back, and a running back doesn’t get credit just for touching the ball 250 times a year.

For each first-round pick from 2006 through 2020, we computed the gap between what the player actually produced in his first four seasons and what his draft slot historically yields for that position. Players who significantly outperformed got labeled a boom. Players who significantly underperformed got labeled a bust. Everyone else landed where you’d expect.

Then we sorted by position.

The data

Look at the safety row. A 17.2% boom rate and a 17.2% bust rate — right in the middle of the pack. Not the safest investment on the board, but not the riskiest either. Perfectly average. More on this later, but to keep you interested, Safety booms: Kyle Hamilton, Eric Berry, Harrison Smith. Safety busts: Mark Barron, Ha Ha Clinton-Dix. One interesting player, that neither a boom nor a bust by the data, is Derwin James. Huge talent, great production now, but it in his first four years, injury concerns plagued him a bit, including a lost season due to a torn meniscus. The old addage, “The best ability is availability” applies here for sure!

All of this might sound like it supports the case for letting Downs slide. If safeties aren’t outperforming their draft slot at a higher rate than other positions, why rush to draft one?

Because look at what the NFL does consider worthy of a premium pick.

Cornerbacks bust at 27.1% — more than one in four first-round corners significantly underperform their draft slot. The boom rate is just 18.6%. That means first-round corners are nearly 50% more likely to bust than to boom. And yet teams draft corners in the top 15 every single year without blinking. Paging C.J. Henderson, please pick up the red courtesy phone.

Edge rushers and interior defensive linemen, the “premium” defensive positions teams are most comfortable investing in, bust at 23.9%. Nearly one in four.

Quarterbacks bust at 25.0%. One in four. Really, this is inflated based on some historic quarterback classes in the 2017-2020 classes

So, the safety bust rate of 17% is lower than corners. Lower than edge rushers. Lower than quarterbacks. The NFL will absorb a one-in-four failure rate at each of those positions without hesitation, but a one-in-six failure rate at safety is somehow too risky for the top 10?

The discount Downs is getting isn’t justified by the data. It’s justified by habit.

The Hamilton precedent

Full disclosure: I’m an Ohio State fan. So you can factor that in however you want. But the argument I’m making works regardless of what jersey Downs wore, because it’s about the market, not the player. And the market already proved itself wrong on this exact question, with this exact profile, four years ago.

In 2022, Kyle Hamilton was the consensus best safety prospect since Eric Berry. Three-time All-American at Notre Dame. Elite size, elite instincts, elite versatility. Most evaluators had him in the top 10 on talent alone.

He went 14th. Thirteen teams passed.

Three years later, Hamilton signed the largest safety contract in NFL history — four years, $100.4 million, with $82 million guaranteed. He’d made three consecutive Pro Bowls, earned two All-Pro selections, and been ranked as the ninth-best player in the entire NFL by Pro Football Focus. His general manager called him “a unicorn.”

Thirteen teams had the chance to draft a unicorn. Every single one of them let positional value math override what the tape was showing them. I’m certain the Giants would love a Delorean with a flux capacitor right now since they passed on Hamilton twice. The Ravens got him at a discount and are now paying him like a top-10 player at any position, because that’s what he is.

Would anyone be surprised if the J-E-T-S took Bailey or Reese at two and then Ty Simpson at 16. Another New York team passing on a can’t miss prospect. Supposedly, Downs won’t fall past the Cowboys, but as DJ previously said, “Weird stuff happens in the safety position.”

The Track Record

The Hamilton story gets the most attention because it’s the most recent. But it’s not an isolated data point.

Eric Berry went fifth overall to the Kansas City Chiefs in 2010. The pick drew the same skepticism Downs is getting now: same arguments, same position, different decade. Berry went on to make five Pro Bowls, earn three All-Pro honors, and battle back from Hodgkin’s lymphoma to become one of the most beloved players in Chiefs history. By the yardstick this analysis uses, he’s one of the clearest booms in the entire dataset: a top-five pick who delivered top-five value over four seasons and well beyond.

Harrison Smith is the quieter case. The Vikings took him 29th overall in 2012, the kind of pick that doesn’t move the needle on draft night. Smith went on to become arguably the best safety of his generation, earning multiple Pro Bowl selections across more than a decade in Minnesota. The model classifies him as a boom at pick 29: he significantly outperformed what that draft slot historically produces. The late-round discount made the surplus value larger, not smaller. Teams that waited for the position to come to them got exactly what the data predicts. Oh, by the way Harrison Smith is close to retiring, do you think Minnesota will hesitate to draft Downs if the Jets pass on him twice?

Hamilton at 14. Berry at 5. Smith at 29. Three different eras. Three different draft slots. One result.

The metric is also honest about the misses.

Mark Barron went seventh overall to Tampa Bay in the same 2012 draft as Harrison Smith. He never justified the slot as a safety, struggled in coverage against NFL athleticism, and eventually converted to linebacker with the Rams. The model classifies him as a bust: a top-10 pick who meaningfully underperformed what that slot should produce. He’s real, he counts, and he’s part of why the safety bust rate in our dataset sits at 17.2% rather than zero.

But look at what Barron’s case actually illustrates. The concern wasn’t that he was a safety, it was that he was a below-average athlete at the position, asked to do things his athleticism couldn’t support in the NFL. The concern with Downs runs in the opposite direction entirely: too small, too short-armed — measurables that ignore three years of elite production against SEC and Big Ten competition with athleticism that has never been the question.

The data doesn’t say every safety is a safe pick. It says elite safeties are. The consensus already agrees on which category Downs belongs to.

The prospect

We’ve talked a lot about how we can learn from the past to predict future successes or failures, but let’s talk now about the actual player in question. Caleb Downs played 44 career games across three seasons at Alabama and Ohio State, and started every single one of them. He finished with 257 tackles, 16 tackles for loss, six interceptions, 12 pass deflections, three forced fumbles, and two punt return touchdowns. He led Alabama in tackles as a true freshman, the first in program history, by the way. I have heard that Alabama has a pretty good track record in football. So, to be the first to do it at that school may be a better data point than arm-length. He earned SEC Freshman of the Year, then transferred to Ohio State after Saban retired and earned unanimous All-American honors in back-to-back seasons. He won a national championship with the Buckeyes in 2024. He won the Jim Thorpe Award as the nation’s best defensive back in 2025, along with Big Ten Defensive Player of the Year and the Lott IMPACT Trophy.

And if that weren’t enough, how about this from Nick Saban who knows a thing or two about coaching Defensive Backs, “This guy is one of the most complete people I’ve ever had the opportunity to coach. He’s a great person. He’s a great student. He’s a phenomenal football player — his competitive IQ, how he responds on the field and reacts instinctively, It’s in the top tier of all players of all time.”

He comes from an NFL family. His father Gary played running back for the Giants, Broncos, and Falcons. His brother Josh catches passes for the Colts. His uncle Dre Bly was a two-time Pro Bowl cornerback.

Multiple evaluators have called him the best overall player in the 2026 draft class regardless of position. John Harbaugh — who coached Ed Reed and Kyle Hamilton — said he’d take Downs “in a second” and used the phrase “potential future Hall of Fame safety.”

And despite all of that, the current consensus has him sliding to 10, 11, or maybe even 12. Why? Because safety isn’t just the position name, it’s what NFL teams are after.

The real question

This isn’t about whether safeties are secretly the most valuable position in football. They’re not. On average edge rushers and offensive tackles carry more structural importance to winning games in most schemes. The analytics community isn’t wrong about that.

But “on average” and “most schemes” are doing a lot of heavy lifting. The question isn’t whether the safety position is worth a top-10 pick in the abstract. The question is whether this specific safety, a prospect that a Hall of Fame coach says he’d take “in a second”, deserves a 5-to-10-pick discount because of a label.

The data says the risk at safety is no different than the risk at positions teams draft in the top 10 every year. The bust rate is 17.2%, right in line with the field and meaningfully lower than most of the so-called premium positions. This isn’t a gamble. The discount isn’t a risk premium. It’s a tax on the position name.

Teams picking in the 5-to-12 range this April have a chance to do what the Ravens did in 2022: draft a generational safety talent at a price created entirely by positional prejudice.

Hamilton went 14th and became the highest-paid safety in history. Every team that passed on him now wishes they hadn’t.

Nobody here is saying Downs will be the best player in the draft. What we are saying, he might be the most underpriced player in the draft, and the market is penalizing him for it.

All this, and we haven’t even gotten to what the data says about the front seven, where the positions teams feel most comfortable spending draft capital carry risk profiles that would make a stock analyst wince. The defensive side of the board has some surprises in it. Coming soon!

Data via Pro Football Reference. Analysis and code at github.com/merrittocratic/nfl-draft-model.

Since the late 1990s only one quarterback has “hit” in the draft with 15 or fewer NFL starts. I think we can all agree Ty Simpson is not Cam Newton…

Here we go again!

A quarterback class everyone agrees is weak. One clear QB1 at the top, or in other years, no clear top QB prospect. Then comes the prospect who’s being talked into the first round on being a coach’s son, arm talent, and the simple, desperate math of NFL teams that need a quarterback more than they need to be right.

In 2022, that prospect was Kenny Pickett. In 2026, it’s Ty Simpson.

Daniel Jeremiah went on The Rich Eisen Show last month and said flatly: “I like Ty Simpson more than Kenny Pickett.” Adam Schefter told Pat McAfee he’s confident Simpson will be a first-round pick. The narrative machine has narrowed its focus and the draft-industrial complex is dialing up the pressure.

But here’s the thing about narratives, for some reason we’ve decided they don’t have to survive first contact with the data.

It’s Merrittocracy time!

The table nobody wants to see

The argument for Simpson as a prospect usually starts with the scouting report: elite processor, football IQ, pocket manipulation. It never starts with the numbers. There’s a reason for that.

Since 2010, four quarterbacks have been drafted in the first round with fewer than 15 career Division I starts. Kenny Pickett wasn’t technically in that group — he had 52 career games at Pitt — but since everyone wants to bring up Pickett, we’ll keep him in the conversation. Here’s the full picture, including the only low-experience quarterback who actually succeeded in the NFL:

Simpson finishes last in every production column except interceptions — which just means he threw fewer passes. He’s the smallest prospect in the group by a significant margin. He has the least prior experience. He’s the only one without a single major award.

You could probably get away with saying Simpson had comparable TD pass numbers to Newton, but look at the rushing line. Full stop, comparison over!

Cam Newton produced 1,473 yards and 20 touchdowns on the ground in 2010 — numbers that would lead 109 out of 120 D1 schools in rushing. That’s right, only 11 schools in all of D1 had a running back that rushed for more yards than Cam Newton in 2010. He was the first college quarterback ever to average 200+ passing yards and 100+ rushing yards per game in a season. He was a one-man offense at 6-6, 250 pounds, producing 51 total touchdowns and 4,327 yards of total offense. He won the Heisman in a landslide, won a national championship, and did it all with physical tools that come along once a generation.

Simpson rushed for 93 yards all last season. On a team that ranked 125th out of 136 FBS programs in rushing.

The production gap isn’t a gap. It’s a canyon if you’re looking for someone who can break the 15 or fewer starts trend!

What “limited experience” actually looked like when it worked

There’s a talking point floating around that Newton proves low-experience quarterbacks can succeed, so why not Simpson?

Because Newton’s “limited experience” wasn’t actually limited. Before Auburn, he spent a full season at Blinn Community College, where he threw for 2,833 yards and 22 touchdowns while rushing for 655 yards and 16 more touchdowns, winning the NJCAA national championship. He arrived in the SEC having already quarterbacked a team to a title. His FBS start count was 14, but his competitive reps told a very different story.

Newton then produced one of the greatest individual seasons in college football history, and still had to be a 6-6, 250-pound generational physical specimen on top of all that production and all that experience for the profile to translate.

That is the minimum viable product for a low-experience quarterback to succeed in the NFL. Simpson is a 6-2, 208-pound pocket passer. He is not Cam Newton.

And when great production wasn’t enough

If Newton shows what it takes for this archetype to work, Dwayne Haskins shows how it fails even when the numbers are historic.

I loved Dwayne Haskins at Ohio State. There’s video of him at 10 or 11 years old walking around the OSU facility wearing a number 7 jersey. He was always going to be a Buckeye. I just wish he had stayed another year. Who knows what might have happened and how things could have gone differently for him….

In 2018, Haskins completed 70% of his passes for 4,831 yards, good enough to lead the entire FBS, along with 50 touchdowns and 8 interceptions. He set or tied 28 Ohio State records and seven Big Ten conference records. He won the Big Ten Championship, earned conference Offensive Player of the Year, and finished third in Heisman voting. Fourteen starts, one year, one of the best single starting seasons by a college quarterback in recent memory. To put that into perspective, Haskins threw for 11 more TDs and 1,000 more yards than Drew Brees’ best season at Purdue!

Washington took him 15th overall. They released him after two years. He threw 12 touchdowns and 14 interceptions in 16 career NFL games. The most productive one-year starter from the pocket in Big Ten history couldn’t make it work.

Simpson’s single-season production isn’t in the same universe as Haskins’s. If that résumé couldn’t overcome limited starting experience, what specifically about Simpson’s lesser version is supposed to project differently? Does Kalen DeBoer have an amazing track record of producing high-level pro QBs? The Falcons might take issue with that statement. If it’s Simpson’s off-platform throws, I think I hear Zach Wilson calling.

The other low-start first-rounders since 2010 fare no better. Trubisky (13 starts at UNC, 2nd overall in 2017) was out of Chicago in four years and is now a career backup. Richardson (13 starts at Florida, 4th overall in 2023) has completed 50.6% of his passes over three NFL seasons and spent 2025 as a backup. That’s a 75% bust rate in this comp group — and every late-round success story from the past decade (Purdy, Prescott, Hurts, Minshew) had extensive college starting experience. This profile doesn’t produce NFL starters. Period.

The tape got worse

Even within Simpson’s single season, the trajectory points the wrong direction.

Through his first nine games, Simpson looked like a legitimate first-round talent. He was completing around 67% of his passes with 21 touchdowns and a single interception. Alabama was rolling. The hype was earned.

Then defenses adjusted. Over his final six games, Simpson’s completion rate dropped to 61%. He threw four interceptions. His yards-per-game average fell from 266.7 to 158.3. Alabama’s offense stalled, and Simpson didn’t have the experience or the counterpunch to respond.

One season of tape is already a thin evaluation sample. When a third of that season trends sharply downward, you’re building a first-round projection on roughly nine games of good football. Haskins, for comparison, didn’t have a late-season regression — his tape held up from start to finish, and he still couldn’t translate it to the NFL.

“But he processes better”

The case for Simpson over everyone in that table rests on subjective scouting evaluation. He reads defenses quickly. He grew up as a coach’s son dissecting film at the kitchen table. His pre-snap processing is elite.

These things may all be true. Processing speed matters for NFL quarterbacks, and Simpson may genuinely be a sharper processor than Pickett or Haskins were coming out of school.

But processing is not a measurable. It doesn’t show up in a box score. And it’s the kind of trait that draft analysts lean on when the production doesn’t support the grade they want to give. “The stats don’t show it, but trust me, you can see it on tape” is the most dangerous sentence in draft evaluation, because it’s unfalsifiable in the moment and only accountable years later, when nobody’s checking receipts.

It’s also worth noting that Simpson and his most vocal media advocates share agency representation through CAA. That doesn’t make anyone’s evaluation wrong, but it’s context that consumers deserve when weighing the hype cycle.

I know ALOT of people now think the combine is useless. You don’t play football in your underwear! But it can be instructive if you do some head-to-head comparisons. If you want to be objective, go watch Ty Simpson and Dwayne Haskins throw back-to-back. In particular, watch the deep ball throws. Ty is 10 yards shorter in air distance and does not at all have the prototypical “drop it in the bucket” catchable ball like a Russell Wilson in his prime.

The smarter play

One last thing to add more clarity, there is not a SINGLE success story of a low-experience QB coming from later rounds to become a successful franchise QB either, so just dropping to a later round to draft Simpson still might not be the play.

So, here’s where this moves from “why not Simpson” to “what to do instead.”

If your team needs a developmental quarterback, the data overwhelmingly says: don’t burn a first-round pick on a one-year starter in a weak class. That’s the Trubisky play. That’s the Haskins play. We know how they end.

Top backup quarterbacks now cost $7–8 million annually on the open market. There’s real value in drafting a developmental QB on a rookie contract — but that’s a Day 2 or Day 3 investment, not a first-round one. And this class has two options that the data supports far more than Simpson at pick 20-something.

Garrett Nussmeier (LSU, projected mid-to-late Day 2) has something Simpson doesn’t: a program pipeline that’s producing NFL quarterbacks right now. LSU has had nine quarterbacks drafted since 2000 — tied with USC for the most in the country — including Joe Burrow (1st overall, led Cincinnati to a Super Bowl) and Jayden Daniels (2nd overall, Offensive Rookie of the Year). That’s not a coincidence. The coaching infrastructure, the offensive system, the culture of developing NFL-ready passers — it’s real, it’s recent, and it’s measurable. Multiple evaluators see Nussmeier as a prospect with the awareness and mentality to stay in the league for a decade or more, with room to develop into a spot starter. Bonus, he’s the son of a coach as well! What a deal!

Drew Allar (Penn State, projected Day 2–3) is a different value proposition. He’s 6-5, 240 with elite arm talent — physical tools that Simpson simply doesn’t have. He’s a former five-star recruit with three years of starting experience and nearly 7,500 career passing yards. His stock dropped after an inconsistent 2025 that ended with a broken ankle, landing him in the mid-Day 2 to Day 3 range. Here’s the honest program pipeline read: Penn State is not a quarterback factory. Their last first-round QB was Kerry Collins in 1995. Also, it’s tough to exorcise the ghost of Christian Hackenberg, a second-rounder in 2016 who never took a regular season NFL snap. Penn State produces linebackers and EDGE rushers, not elite passers. But with Allar, that pipeline weakness is already priced into his draft position. You’re getting first-round physical tools at a Day 3 price — and if you’re betting on developmental upside, bet cheaper.

Both options cost a fraction of the draft capital. Both have more experience. And if you’re wrong about either of them, you spent a third- or fourth-round pick — not a franchise-defining first-rounder that haunts you for half a decade.

Use that first-round pick on a premium position player: an edge rusher, an offensive tackle, a cornerback. Positions where the hit rate is dramatically higher and the rookie contract provides enormous surplus value. Or, draft your project quarterback where the cost of being wrong doesn’t cripple the roster building plan. Build the roster first and when it’s close to being capable, draft the QB.

What we’re not saying

Nobody here is calling Ty Simpson a bust. He MAY turn into a quality NFL starter. His processing MAY be everything the evaluators say it is, and the right coaching staff may unlock a long career.

But “projects better than Kenny Pickett” is a specific, testable claim — and the data doesn’t support it. Lower completion percentage. Fewer touchdowns. Dramatically fewer starts. A late-season regression that neither Pickett, nor Haskins (truer lower experience comparison QB) showed. Ninety-three rushing yards in a season where the only successful comp in this archetype rushed for 1,473. The smallest frame, the thinnest résumé, and the fewest accolades of anyone in the comparison. A production profile that finishes dead last in a four-way comparison against every low-experience or one-year starting quarterback drafted in the first round in the past 15 years.

The consensus isn’t wrong because it’s optimistic about Simpson. It’s wrong because it’s making a comparative claim without comparative evidence. That’s not analysis. That’s storytelling.

And storytelling is how you end up drafting Kenny Pickett at 20.

Data via nflverse/nflreadr, Pro Football Reference, and ESPN. Code and methodology at github.com/merrittocratic/nfl-draft-model.

Every sport comes equipped with a narrative-building machine.

The NFL has it’s draft cycle beginning right after the Super Bowl. Every year, analysts begin repeating each other’s evaluations until mock drafts harden into consensus, a team picks the guy everyone expected, and three years later nobody goes back to grade the pre-draft analysis. In the NBA, the narratives center around “Is this team really good enough to win the Finals”, or my favorite, “Who will be the next Face of the League”. In golf, “Who will be the next Tiger Woods?”, and is it their time to win the Masters? Gotta be honest though, right now, being the next Tiger Woods doens’t feel too appealing.

The cycle works the same across the entire sports landscape: narratives are created, audiences consume them, the moment passes, and accountability expires. The experts just move on to create the next cycle.

Merrittocracy exists because that cycle is broken — and because the tools to do better are sitting right there in the data.

What this is

Merrittocracy is a sports analytics brand built on a simple premise: the conventional wisdom should have to survive first contact with the enemy, a man with some code who’s not afraid to use it.

Sometimes the consensus opinion will hold up. The wisdom of the crowd does gets things right, and when it does, we’ll show how the data that backs it up. But when it doesn’t — when the narrative is running on vibes instead of evidence — we’ll show that too.

Why now? Well, two things will become clear with Merrittocracy, I love College and Pro football, and the NFL Draft is the perfect intersection of both. The Draft ia also the perfect proving ground: a months-long evaluation cycle, measurable outcomes, and an entire mediasphere that produces confident opinions with surprisingly little accountability or data. The draft is the starting point, not the finish line. The same narrative-checking lens applies anywhere sports media produces consensus takes that outpace the data. Football, Basketball, Golf, and maybe a little bit of baseball, we’ll cover it all. Sorry, no Hockey…

The draft model

Our launch vehicle is a boom/bust probability model trained on every NFL draft class from 2006 through 2020 — roughly 3,750 players with four years of career data to measure outcomes against.

The model takes the features available before draft night — combine measurables, college production, draft position, age, experience — and asks a simple question: given where this player is being drafted, how likely is it that he significantly outperforms or underperforms that slot?

That’s the key distinction. We’re not predicting who’s “good” or “bad” in a vacuum. We’re predicting who’s good relative to where they’re drafted. A player taken 5th overall has a much higher bar to clear than a player taken 45th. The model measures draft-pick-adjusted production — the gap between what a player actually delivered and what his draft slot historically produces.

From that gap, three labels:

• Boom: significantly outperformed draft position

• Bust: significantly underperformed draft position

• Expected: landed roughly where the pick historically lands

The output is always a probability range. “Our model gives him a 35–55% boom probability” is more useful than a talking head saying “this guy’s a stud.” Honest analysis admits what it doesn’t know.

What makes this different

Most public draft models treat college program as either a flat categorical variable — Alabama = Alabama, regardless of era or position — or ignore it entirely. That’s a mistake.

Our primary novel feature is the program pipeline: rolling 10-year, position-group-specific draft outcomes for every college program. Ohio State’s track record producing wide receivers is a completely different data point than Ohio State’s track record producing quarterbacks. Georgia’s defensive line pipeline tells you something different than their offensive line pipeline.

Why 10 years and not all-time? Coaching changes and scheme evolution matter. Saban’s Alabama produced prospects in a fundamentally different developmental environment than what came before — and what’s coming after. A rolling window captures current program identity rather than diluting it with decades of irrelevant history.

When someone says “trust the Alabama brand,” the program pipeline asks: at which position, in which era, producing what results?

Three algorithms, Seven sub-models, one comparison

This section is a little more technical, but don’t worry we’ll get through it! We don’t just run a single model. We will end up running thousands of simulations across multiple variations of the data. The model complexity will range from functional to frontier, and we’ll compare the results head-to-head:

XGBoost: the gradient boosted tree baseline, tuned across dozens of hyperparameter combinations. If you’ve seen a public draft model, it’s probably some version of this.

TabPFN: a foundation model for tabular data, published in Nature. Zero tuning — a single forward pass produces predictions. If a zero-shot transformer beats tuned XGBoost on NFL data, that’s a headline.

TabNet: attention-based deep learning with built-in interpretability. TabNet tells us what the model focuses on when evaluating a specific player. Being able to tell you “here’s what the model pays attention to when it looks at this QB” is way more interesting than “the model says bust probability of 42%.”

This will be a genuine methodological experiment for the analytics community, and three independent perspectives on every prospect for NFL fans.

Shocking to hear I’m sure, but not every position on the field has the same measurables, so why out them all in one model? Instead of one mega-model trying to predict outcomes for quarterbacks and offensive linemen with the same features, we run position-group-specific sub-models: QB, WR/TE, DL, OL, DB, LB, RB. Each group shares enough evaluation DNA to use common features while keeping enough historical observations to train reliably.

Beyond the draft

The draft is where we prove the concept. The lens is portable.

The same question we ask about draft prospects, does the data support what the consensus is saying?, applies everywhere in sports. Playoff narratives that ignore the numbers. Award races driven by storylines instead of performance. Historical comparisons built on nostalgia instead of evidence.

We’ll go where the narratives are loudest and the data is most interesting. The NFL Draft is April 23. After that, the work keeps going.

Everything is public

The code is on GitHub. The methodology is explained here. The model outputs are transparent. If we’re wrong, you’ll be able to see exactly why — and that’s the point. Accountability is what separates analysis from content.

Why “Merrittocracy”

Because analysis should be earned by the data, not inherited from consensus.

If you had a two-minute elevator ride with just me, you’d find: I flew helicopters in the Navy for 20 years: H-60s, 4,500 hours, eventually commanding a squadron. Then I pivoted into data science, where I’ve spent the years since building production-grade predictive models and wondering quite loudly (ask my wife) why nobody applies the same rigor to the NFL Draft.

Military aviation has a term for when you stop trusting your instruments and start flying by feel: spatial disorientation. It’s the moment your senses diverge from reality, and it kills people. Draft analysis has the same problem, analysts abandon the data and start flying by vibes. Nobody dies, but franchises crater for half a decade.

The draft is full of confident people flying blind, let Merrittocracy be your instrument panel.

Follow along:

• X: @Merrittocratic — narrative-checks, data viz, short takes

• Substack: substack.com/@themerrittocracy — deep dives, methodology, full model output

• GitHub: github.com/merrittocratic/nfl-draft-model — code, data, transparency

First analysis piece drops this week.

Will Ty Simpson even get drafted once this gets out there?

Data via nflverse/nflreadr and Pro Football Reference.