Hope your NFL season is going well. If you are a fan of some teams, it’s been fairly splendid. If you are a fan of other teams, it’s going not so great. And if you are a fan of the Miami Dolphins, it’s going very very bad, my condolences. But this isn’t the space to talk about the Dolphins explicitly (and I’d rather not talk about them implicitly either but…).

Today, I want to run a quick piece on one of my favorite NFL metrics, DVOA. If you are unfamiliar with DVOA, you can read the wonderful primer by it’s creator Aaron Schatz over at FTN (shoutout to all the former Football Outsider-heads too). But as a quick explanation (again from Aaron):

DVOA is a method of evaluating teams, units, or players. It takes every single play during the NFL season and compares each one to a league-average baseline based on situation. DVOA measures not just yardage, but yardage towards a first down: Five yards on third-and-4 are worth more than five yards on first-and-10 and much more than five yards on third-and-12. Performance is also adjusted for the quality of the opponent. DVOA is a percentage, so a team with a DVOA of 10.0% is 10% better than the average team, and a quarterback with a DVOA of -20.0% is 20% worse than the average quarterback. Because DVOA measures scoring, defenses are better when they are negative. Every year is normalized so the average for that year is 0%.

One reason I really like DVOA is that is shares a similar architecture as success rate, but also adjusts for context of the situation, the quality of the opponent, and even at times Aaron and the team will make adjustments for injuries to the team/opponent. It also allows us to compare teams better across years and eras. Like all stats, it’s to be used in combination with others, not as the sole source of truth.

But DVOA can be tricky. Early in the season prior to week 4, there is no opponent adjustment. And while there is some signal in the early noise, it lacks a core feature of the strength of DVOA. Slowly over the season, that opponent adjustment gets layered in by 10% a week from week 4 (40%) until week 10 when it’s reached 100%. That can make it complicated to split results and adjustments during that period, as team performance changes but so does the opponent adjustment. It’s not a flaw in the system, but just something you have to consider in those periods of the first three weeks and then weeks 4-10 (everything after is basically just team performance quality and updating some Bayes-like updates of opponent adjustments).

But surely we don’t have to wait until week 10 to feel good about a team right? And surely in the early weeks there is some gold that can be mined for how well a team is and is going to be, no (spoiler alert: yes there is)?

So that is what we are going to look at today. When do DVOA figures start to bake in some real truth to what the final year end figures will look like, which part of DVOA can we trust a little earlier, and can we predict future DVOA?

I’m really going to keep this one mostly quick and high level, trying not to bore folks here.

Subscribe now

There is a very large historical DVOA archive that has yearly DVOA dating back to the 70s, and weekly DVOA dating back to 2002. Since the latter is more what we are looking for, we are able to back run the analysis from 2002 to 2025.

So what is stabilization? Is it when weekly variance shrinks? Is it when opponent adjustments being layered in don’t add as much as weeks past? Is it when the team’s final result starts getting locked-in? Is it when the current DVOA predicts future DVOA >50%? We are going to look at the last two for this one.

Our first chart here is just simply when the R2 (correlation coefficient) for each of the three parts of DVOA we are looking at (offense, defense, and total - sorry no special teams in this one) hits various levels. The orange vertical lines show when the first opponent adjustment kicks in, with week 10 representing the full bake.

For offense (green) and total (blue), the 50% R2 threshold hits basically right at week 4 when those opponent adjustments kick in. At this point, half of the final offense and total DVOA is explained by the current DVOA. That slope does still continue a bit but at about week 7 or 8, we are over 70% there.

For defense, it’s a little less sticky early on. It’s not until week 6 that it crosses the 50% threshold (offense and total got there two weeks earlier), and not until week 10 when it crosses the 70% level (offense and total got there three weeks earlier). But once we get to week 10, we’ve loaded almost 3/4th of the entire end DVOA for each part.

Why does offense stabilize faster than defense? Across each regular seasons, offenses have a a wider true-talent level dispersion with a standard deviation of = 13.52%, whereas defense is at 9.54%, even though both start with nearly identical Week 1 noise (28.52% and vs. 26.93% respectively). Offensive performance is driven by quarterback play and general scheme continuity, while defense can be heavily dependent on opponent offensive style, high-variance turnovers, and third-down conversions.

This one is a little busier, but what is functionally telling us is when we’ve moved out of small sample size zone (what we are calling the trust weight) and not shrinking it back to league average.

The solid lines (predicting remaining unplayed games) strip out the games already in the bank and asks “if a team is 20% above league average right now, how good will they play over the rest of the schedule?” When a solid line crosses the B 0.50 dotted line, the number starts representing more signal than noise, where real talent starts becoming >50% small sample size noise. Offenses hit this line after week 5, with total after week 6. However again, defense is the laggard here, it’s not until week 13 that that it crosses over.

The dashed lines are predicting final cumulative DVOA. Early games stay on the team’s year-end record, these lines sit higher. After week 4, a team keeps 55-58% of a their total or offensive DVOA when projecting their final year-end number.

Our last chart here is a little more standard, this is simple the mean absolute error through each week in relation to the final DVOA ratings. After week 1, a team’s total DVOA will be ~38% off from where the year end DVOA will be. After week 4 opponent adjustment starting, it’s closer to ~13%, and after week 7, the slope flattens out and it is just getting only marginally more accurate relative to prior weeks. At week ten, any of the three metrics will be only 5-7% off their year end equivalent.

So all that to say, generally, we can start trusting DVOA to start looking like the year end figure for offense and total DVOA at week 4, really feeling good at about week 7 or 8. Defense, we’ve got to wait a bit longer until probably week 7 if you want to be aggressive, more like week 10 to feel better, and maybe even a little later until week 12 or 13 to feel really comfortable.

Subscribe now

What if too we can make some predictions on DVOA? We know the shape of the curves for each week, how can we tell competitors from cons? Frauds from faithfuls?

If a team starts 3-0, with a total DVOA of 30%, 20% ODVOA, and -15% DDVOA, we can estimate that their future metrics will be:

Total DVOA (+30.0%): Expect +8.7% ROS (29% * 30%) and +12.4% final DVOA (+/- 13.8%).

Offense DVOA (+20.0%): Expect +6.9% ROS (35% * 20%) and +9.2% final DVOA (+/- 10.0%).

Defense DVOA (-15.0%): Expect -2.5% ROS (17% * -15%) and -4.7% final DVOA (+/-8.0%).

But when week 4 and those opponent adjustments kick in and they are still stable at those same levels, it looks a lot more clearer

Total DVOA (+30.0%): Jumps to +12.4% ROS (41% * 30%) and +16.5% final DVOA (+/-12.6%).

Offense DVOA (+20.0%): Jumps to +9.2% ROS (46% * 20%) and +11.7% final DVOA (+/-9.2%).

Defense DVOA (-15.0%): Moves to -3.9% ROS (26% * -15%) and -6.5% final DVOA (+/-7.5%).

So if we make some forecasts based on what we know historically after week 3 and the current DVOA data for 2026 teams, we get some ROS projections

Nothing is too wild here, no team makes a huge jump/drop. The 49ers are forecasted to stay the top team due to how strong their offense has been, which if it survives the week 4 opponent adjustment it will only solidify this forecast.

A big standout is Dallas, who has been the #12 team by DVOA, driven by the 5th best offense, however they have been the 27th best defense. The DVOA forecasts expects them to be more like a top 5 team for the rest of the season, as it sees heavy regression in their defense. You may push back on this one and I understand, this DVOA forecast doesn’t know about the 2025 Cowboys and how bad the unit has been for awhile now (though they just traded for Joey Porter Jr).

Also the Minnesota Vikings look shaky here. Currently the #8 team by DVOA, it’s driven entirely by their #1 ranked defense instead of their #30th ranked offense. And while Brian Flores and his units are fun to watch and generally a nightmare of offenses, we again know it’s not until mid-way through the season that we can start trusting those defensive figures. For instance, last year through three weeks the Packers were 3rd in DDVOA and finished 19th (in part due to a Micah Parsons injury). Or even the opposite, the Texans last year started out 0-3 with the 19th ranked defense before finishing the year with the 2nd best. Another one you can be skeptical of though, as the forecast doesn’t know that Brian Flores is leading the defense and may be immune to regression to the mean.

And similarly we can see the product of those shrinkages and teams that look a bit more buy low vs sell high mirages

But we don’t necessarily need to do our own forecasting, FTN does a version already called DAVE (DVOA Adjusted for Variation Early) and blends in preseason forecasts week by week to current results.

We can compare our forecast to the current DAVE forecast and see exactly what we were talking about. DAVE knows that the Vikings defense is ran by a madman in Flores, and it decidedly does not regress the Vikings defense closer to the mean. Also adjustments to teams like Arizona (not starting Carson Beck), the Packers (missing Parsons temporarily), and the Giants (without Dart).

Even when we test our shrinkage model and add in some preseason DAVE, it’s actually not far off from what current DAVE is already getting.

So I would say for forecasting, DAVE is probably the best way to go here if you want to include some priors. But you can also start believing in the non-DAVE DVOA figures after a few weeks depending on the metric.

One last question: how predictive is prior year DVOA to the next. It slightly depends on the era (pre-salary cap and not), but total DVOA, offensive DVOA, and pass offensive DVOA are fairly sticky, at 20-25% R2. I know those seem like low figures in an absolute sense but for sports analytics, that is a nice r-squared.

I like that last graph a lot because it’s answering the question of “at what week in the current season does a team’s performance become more predictive of their future play than what they did all of last year?” Again like we saw with other measurements, offense and total DVOA stabilize really quickly, in week 3-4 we see the in-season performance cross the green dashed line, which is our r-squared of 25%ish we just talked about. Defense again takes a little longer, more like week 5 or 6.

Or to say, until about week 3 or 4, offense and total DVOA from the prior year is more predictive of where the team will end up in the current year for those two metrics, which then current season performance so far becomes equal or better at predicting. Defense as usual takes longer. The curves start to flatten out a bit, especially defense, because there are fewer forward looking games remaining, so there is some sample size noise.

Subscribe now

I’ve got a self-appointed peculiar beat, being the purveyor of “I spend money on public analytic sites so you don’t have to” to hopefully give folks a taste of what a site offers so they can decide if they want to subscribe or not (though you should still, my opinion is generally wrong).

I’ll be honest gang…we’ve been in a slump. Here is my original piece on this topic from 2023, then the follow up from 2024. From 2023 to 2024, we had real change and product launches and as I said at the time, shit has pivoted.

But from 2024 up until about a few weeks ago, shit stopped pivoting. There were some small (not trying to be pejorative here) launches and tweaks, but nothing what I would call major came out.

But as Eagles fan Kevin Malone said (sorry Kevin, the Eagles did not clinch the NFC East in 2008 or 2009 like you had hoped), “it’s just nice to win one.”

Not only did we get a major product launch, we got several, a half dozen or more. And not more than just tweaks only, but really completely new and different tools to analyze football with.

We’ll go one-by-one a little bit below, but first I want to say the way I think about these products is that being unique is a good thing. Obviously there will be overlap in data, there is only one way to define red zone score% or who got the most first downs. You can get murky on coverage shells and run types and where air yards begin from, but for the most part, we’ve got a good stable of standardized data points that are just really table stakes at this point.

But as a data provider, you have to ask yourself if what you have is unique? Or to the point, why should I pay for your product instead of someone else’s? Price alone can’t dictate that, as a $20 product and a $200 product aren’t usually the same. And that’s not to say you can’t have a good $20 product or that the $200 product is better because of the price. But data providers need to ask themselves what am I giving my subscribers for their cost, how is it different, and is the cost right?

And at the same time, you as the consumer, you can set your overall budget on how much you want to spend, but don’t settle for a single solution if you can afford not to. There are genuinely unique (forgive me for modifying the word unique) features across various sites, and nobody has everything (even if they tell you they do, they don’t). And don’t assume that more money = more product. There’s a product suite out there that charges $800 a year and it’s on par with some products that cost you only $50 a year.

Allow me just a few more inches of webspace to rant, err well, post a screenshot of my tweet:

But data sites, no matter the sport, show don’t tell. Show us the data and allow us to access it, don’t just post a graphic or mention a stat on the show. And most importantly, don’t show things in a promotion of your product that your product does not allow the user to do! PFF was guilty of this for years, they’d talk about QB accuracy charting or coverages or WAR on podcasts. And the worst offense was when they would post screenshots in their consumer articles of data from a product you cannot buy, yet planted deadcenter in the article is a “subscribe to PFF Elite/Edge/+" as if you could access that paywall.

I have one bit of disclaimer before we walk site-by-site: I am not a journalist but I do try to be ethical. I am just a dumb finance guy, who someone should have never shown me the URL to FanGraphs, which kicked off my love for sports analytics. My weird beat and corner of the universe I have imposed on myself is tracking these types of sites. As I interact with people on Twitter and here, I did have people who run some of these sites offer me either early access to a beta site, a discount to a subscription, or a free subscription. I never asked for it, it was proffered. I made clear that while I do try to not trash hardworking people’s work, I was going to be fair in my analysis and writeup, that I will treat them like I was any paying customer. This post is not sponsored by any site, and no site made condition on access to the site being that I write them only in a good light. And to be clear, I paid with my own hard earned US dollars for access either in full or in part to many of the sites. Though to be honest, I did find a promo code online for a few places to get a discount too but I think that’s fair game.

Okay let’s talk about websites. These are solely in order from when their newest product was launched or at least when I heard about it.

FTN Stats IQ

https://stats.ftnfantasy.com/

I’ve been a fan and friend of some of the FTN folks for a few years now, and I’ve always been a big believer in DVOA, so when FTN and Football Outsiders merged awhile ago, it was win-win.

They rolled out StatsHub and it was a pretty big stepchange for analytics. I think at the time, it was the only site really with filters and splits, being able to slice by things like coverage schemes, downs, and distances.

Now they have their new version of it, named Stats iQ that includes Iverson (AI - get it?), as well as a bit cleaner and smoother grid. If you are familiar with StatsHub, then the navigation will be about the same. You can pick your position, a tab for the data criteria you are looking for, then filter from there.

They are also one of the few sites that has offensive line data, included charting pressures and quick pressures. I think it’s the only site that charts time to pressure

Like many other sites you’ll see below, they’ve got a chat bot to help answer questions through data.

Let me stop for a minute: I am not an AI expert. I use Claude and Gemini for various projects, but that is about where my knowledge stops. I am not qualified to judge an AI model on how well it does, fast it responds, or accuracy other than spotting an obvious inaccuracy. I’m sure many of these models are good, this isn’t the space to battle that out.

Though the response time was quick, I couldn’t quite get it to answer my question in the first try or two:

But otherwise the product is a nicer and cleaner portal that StatsHub (which was good to begin with).

Why you might need it: DVOA to me is a core product itself, something I’ve been more than happy to pay for for years now and will continue to do so.

What is it missing?: It’s a very streamlined offering, which is a compliment and a criticism. It’s not flashy, but it includes a good bit of details for every position, you can export most pages, and the filters work. You can’t do multi-season.

Price: Base StatsIQ starts at $109.99, but that includes DVOA as well. They have additional tiers for play-by-play data and more enterprise level pbp (not covered under the scope of this Substack).

Fantasy Points Data Suite 2.0

I hate to use this word, I really do. But as I tried out the site (and I was very familiar with the 1.0 product), I couldn’t help but keep thinking of one word to describe it: miracle. It’s miraculous. A product sent down from a data god after a few years of toiling in the wilderness of calculating our own EPA like Paleolithic Excel dwellers and only being able to split out performance by down, distance, and quarter.

If you aren’t as familiar with this company, they feature a lot of ex-staffers from various fantasy and data sites that you are familiar with. When you hear them talk about their site, they don’t hide that their goal is to take on those companies. They believe they have the best data, product, and it’s at a fraction of the price that teams and enterprises pay for something reasonably equivalent.

Just at the simplest level, the stats grid from 1.0 carries over a lot of the same layout and customization, but the filters go further, as does the data display. For instance, you can split out or filters players by who the head coach or play caller was, their injury status, their age, combine data, draft data, college, experience, and with new views on how it is displayed by things like percentile, rank, and per game.

For instance, let’s look at EPA per target, minimum of 150 targets, for a first round receiver, over the past three years.

The EPA/Tgt column wasn’t in that spot initially, you can drag and reorder columns. As well as create custom columns. Here is EPA per Zone rushing attempt (and by the way you can further set minimums on your custom columns too).

I believe for the first time anywhere, you are able to do multi-year selections for data. This has been a personal gripe for me for years now on why this was never available. I know multiple enterprise products that did it, but their consumer facing side did not and it didn’t really make any sense. And here with Data Suite 2.0, it’s doable across pretty much every page.

Their leaderboards have always been great, but I want to speak more about some of the new specific features I like. First, let’s start with players pages. They include a large amount of stats and splits and graphics that includes routes targeted, heatmaps that are splittable, and even an injury history as a nice touch.

Then they’ve got something really neat in coach and play caller pages

Both at the leaderboard level above, but also for detailed coaching level. The offensive coaching pages are a little more detailed than the defensive side (and it would be nice if you could search for a coach in the search bar), which makes sense given the site is mainly geared towards fantasy, which leans towards the offensive side of the ball (sorry IDP players but it’s true).

On to probably the biggest feature (in my opinion). Let’s step back to me at my computer 6-8 years ago. I’m deep into football analytics and I’m trying to understand how to write code in R but it’s really like learning a new language that only a small group of people speak (relatively speaking). And if you don’t know the answer, it’s not like someone can just say “oh you don’t pronounce the h in that word”. It’s “can you paste your reprex?” and then you’ll have maybe 5-10 different solutions or packages you can try and the resolution isn’t usually “you forget a pipe” but something deeply technical. And all I wanted to do was make a scatterplot with NFL logos, which required the data (pre-sorted or you can write more code and sort yourself), the URL to the logos, and the code.

Flash to present day, you can just do that and more in the Data Suite tool itself.

And it’s not just scatterplots, but about any time of infographic you want, over many timespans

These are all just nicely done, out of the box, filterable and splittable, and customizable.

I mean, you’ve even outcomes by lineup combinations, a page for the OL, skill positions, defensive front, and defensive backs (and again, these columns are sortable, splittable, filterable, and can create custom columns).

They even recreated Stathead but with advanced data. You can look for individual games and for seasons.

So here is QB games in 2025 with an ADOT of <=5 yards, YAC% of 80% or more, 20+ dropbacks, and the team won.

Receiver seasons with an EPA/Tgt of >=0.5 and 100 targets

I’m going to speed run through a few things here:

Shadow reports - here is Pat Surtain vs Ja’Marr Chase

Game previews

Game results and box scores

Team pages

An injury database you can split out by injury and see the performance for the players that week too

On/Off splits

I simply cannot walk through every feature in this piece, it would probably need it’s own entry, and at that point I suggest just buying the product. There is a mini-walkthrough from the team themselves that covers a lot more ground than I did, but just wanted to include some of the things that wow’d me. They really just thought of about everything and it’s in a really nice presentation and visualization.

I also know that they have hired additional staff for future features. They are hoping to have soon some more OL/DL data, and possibly over the next two or three years start rolling out college data once Data Suite 2.0 is fully tested out over the NFL season.

Why you might need it: If you want highly detailed, sortable, filterable, splittable metrics, great data viz, custom visualizations, and detailed charting data.

What is it missing?: It’s hard to nitpick this one. It’s still fantasy focused, so the lens of stats will generally be viewed through the offense. There is some defensive data, but the player pages on offense are a bit deeper. There doesn’t seem to be any special teams, kicker, or punter data.

Price: The actual data suite is $200, but it’s a little tricky to get there. There isn’t a way to add just Data Suite to your cart, you have to also add one of their site subscriptions too and then remove the selected non-DS subscription from your cart. Though I will add, I’m always for supporting analytics sites and I don’t have a take on how well their non-DS products are, but if you can throw an extra $60 their way I’m sure it will go to good use.

PFF

If you know me, you know that I have been a PFF acolyte for years. I really do fully buy into their grades and process. I think they do a great job of thinking about the game, providing detail, and quantifying things. I’ll always have a PFF subscription because I value their data. But that doesn’t mean I’m not critical of them, and I think in my first entry in this series back in 2024, I was plenty critical and voiced my frustrations. I can say two things: I know that feedback was seen and heard, but the feedback (not just from myself but others) wasn’t fixed quick enough.

The Stats 2.0 product sat mostly dormant feature-wise for years. It wasn’t until the acquisition by Teamworks that allowed enterprise to split from consumer. Previously, most resources at the company went to the enterprise side and that makes sense, that’s where the money was. But now that the two are split, the remaining PFF organization can focus on the consumer business. Though they can’t ignore enterprise of course, that still remains where the majority of the Teamworks data must go, so there will always just be enterprise data points that PFF consumer cannot provide.

I am happy to say that after talking with some folks, the PFF (just assume from here on out I am talking about the consumer business) is dedicated to creating a great consumer product. They’ve already revamped their website and launched what they are calling PFF Pro. What this new tier gives you is access to the PFF Pro portal (note: with the lower tier subscription, you can still access Stats 2.0 but not PFF Pro), which is for the scope of this piece where there is a big upgrade over Stats 2.0.

Another one here where I won’t go through EVERY new thing, but some important features include things you already know with Stats 2.0 but additions.

The game cards for past games and previews for upcoming games are now tucked away nicely and cleanly (you may remember the preview card being at the top, the game card being in Stats 2.0).

Team reports feature not just team specific data but leaderboards that compare the selected team to the rest of the league. You can change this by offense and defense, which reveals more pills to modify the team leaderboard.

And most of these (all?) can be turned into a scatterplot quickly and cleanly

And something I really love that might just apply to me: you can now just see the entire offense and defense as a whole. With the old reports, on offense you can to go position by position (defense you could do as an entire side). So now you can just export all offensive player grades (both at the season and game level).

Also on team pages there is depth charts (not sure this was in the older PFF site). You always had the visual of offense/defense you could adjust by personnel and opponent, but I don’t think I remember a depth chart. Is this huge? No, but it’s a minor touch that shows they are thinking of enhancements.

The position reports dashboard is closely the same, but with now a displayed top 5 for each report type.

And the report view may look similar, because it is, but it now includes additional columns. There are too many data points to list that were added, it’s nicely in the hundreds between players and teams (with more to come).

New notable additions include:

EPA splits for QBs

The RB page is technically rushing report, but it now includes receiving grades for RBs too

A new and exciting one that just rolled out a few hours before I am writing this (which is why I wanted to wait for all these sites to get further along), we now have three new datapoints for offensive linemen.

Pass block win rate (for all pass sets and true pass sets)

And positive and negative run block play rates

As of now, those are the main individual player additions. I fully expect more to come, this is just as of NFL kickoff.

I’m also happy to report the CSV export works as intended! I know that also seems like a minor detail, but when you export a page, all your filters stick. So when you export passing grades, you just get the QBs rather than anyone who threw a pass that year.

You might have noticed in some of these screenshots the new scatterplot builder tool, which is a welcome addition.

And of course since it’s PFF, you’ll have college data too for this (and all the previous reports).

For the most part, anything you can view on team and player reports, you can turn into a scatterplot.

Player pages now include contract data too.

On the topic of contract data, there is a feature that isn’t live but still in testing. They are at some point going to have something similar to the OverTheCap contract calculator many are familiar with. Allowing you to see what the cap impact would be for cutting player or signing a free agent (though unlike OTC you can’t do things like restructure, post-6/1, and the free agent contracts are non-negotiable/modifiable).

There is also connectivity now to PFF’s data via API. I’m not going to get deep into this one because it’s 1) beyond my skill level and 2) I haven’t had any time to try anything out. But on its surface, users are able to directly pull many of the data points on the team, position, and player reports through terminal connectivity.

. There are plenty of data points to pull, and for advanced users it will save you a step of having to manually export each facet.

They’ve also nicely created a community page, allowing users to upload API use cases for others to view.

Like others as I have said, they also have an AI chatbot, known as Guru. In testing, I’ve had good and bad responses.

Sometimes it was quick and accurate

It’s been clear about when you are asking a question a user doesn’t have access to the data for, which means Guru also doesn’t have the data for.

But also it sometimes does give a weird result. Like when I asked for PFF passing grade for 2018 to 2021 combined, it took that literally and added up all the passing grades.

Or when I tried rephrasing it, I think it got the grades wrong (I’m fairly certain Mahomes has a higher grade than that since his 2018-2021 seasons went 92.3, 86, 90.5, and 75.7 and Russell Wilson had lower). Also I’m pretty sure Tom Brady didn’t take 45,165 snaps in a four year span nor did Justin Herbert who was drafted in 2020 take 23,816 snaps in two years.

Why you might need it: If you use PFF grades and want detailed OL/DL/CB data. For those familiar with the Stats 2.0 environment, this is an improved version. If you want college data, you really have no other alternative. And not only is the data there, but it’s as high of quality and deep as their NFL data.

What is it missing?: Is it everything I wish it had? No, no tool is. It doesn’t have multi-year selections and there aren’t any filters beyond just draft year, position, and team. So if you want to know which QB had the highest PFF grade on 3rd down or which receiver had the most targets against cover 2, you cannot get that data as of now. There’s no PFF WAR (which was always pie in the sky to be fair), special teams data remains the same, and charts are limited to just scatterplots.

But it’s a massive improvement over Stats 2.0 and it at a minimum catches back up to where the competition is. And they have released a new update every single day since the beta went live (obviously not a pace they will continue but it’s proof they are trying to provide a dynamic product).

Price: Everything above is for PFF Pro, which comes with a $199.99 cost. It’s a genuine improvement over Stats 2.0, which costs $99 annually. To me, that improvement is there for an additional $100 and I’m assuming all future product rollouts will only come to those with a PFF Pro subscription.

SumerSports

When I start thinking about a product I am going to pay for, the starting point for me is a simple question of “what does this site do that is unique?” For better or for worse, we are no longer just trying to find which site splits passing stats out the deepest. Now we’ve got websites that have truly unique features, meaning that you actually have more choices to stretch your budget.

SumerSports new SumerPass has a unique feature you just literally cannot get anywhere else and we’ll start with that.

As I type this and the Patriots go up a touchdown in Seattle on opening night, SumerLive lets you consume the game in a completely different way. No need to wait until the next day or two for charted data, EPA tables to update, or grades to post. Almost instantly you’ve got a detailed breakdown at the game level, team level, and player level.

I can tell you that so far New England has run nickel defense on 100% of their defensive plays so far.

Their average time to pressure on the Seahawks is 2.4 seconds

And Will Campbell and Jared Wilson have both allowed a pressure, with Campbell’s at 2.67 and Wilson’s at 3.12.

This exists nowhere else. And in my opinion, it won’t ever anywhere else for the foreseeable future unless the league itself creates a product under NFL Pro (I doubt it). And this is also an incredible product that not just is neat, but also enhances knowledge and how we consume the game both in real-time and in review (you can find an archive of every game dating back to the playoffs last year plus every game going forward as it happens).

Sumer has player stats, and before you ask “how many times can sites split passing stats…” and skip this section, there are some nice details in here.

First you’ve got leaderboards that are filterable/splittable.

(editor’s note by me: after an update, you can do multi-year stats on these leaderboards!)

There are only about a dozen positions and roles that are currently live, so I can’t dive into all the details on the leaderboards, though I’m optimistic these will be well done when they do roll out. Just earlier today a few of these were released, and a source says they hope to have them all done over the next week or so.

Same sort of leaderboard for the team level

There are also player pages (there are no team pages). A lot of data points you can find at other sites too, but I think there are a handful of Sumer only stats as well. I’ll add that there doesn’t seem to be a way to search for a player outside of just going to the positional reports and searching for the player. So to find a RB you want to view, you have to actual go to the default QB page first then switch to RB then search. Just having a search button at the top right would be better.

And again, Sumer has their version of an AI chatbot in SumerBrain. It has actually been out for quite a while (maybe a year or so?), and I’ve used it a good bit in the pre-SumerPass rollout. It works good, as well as you could hope and it handled some tough questions well.

It does a good job of both returning a response but also providing context on the response rather than just shouting the numbers at you.

Why you might need it: As I said, SumerLive is a product you cannot get elsewhere and I could even see paying for just that, getting all the extra features for “free” in a sense. They also do have good breakdowns on player pages and I’m hopeful for more stats when they roll out more positions and roles.

What is it missing?: As described, some position/role data is still outstanding and I can’t comment on if it’s good or not since I haven’t seen it. It’s possible it’s more nice than great. The tables are not exportable and there is no real way to pull data from SumerLive - maybe that isn’t an issue or something they want to offer, but it would be nice to do.

Price: You can pay a few different ways for SumerPass, be it weekly, monthly, or annually. Again it’s not really for me to decide, but I think $100 is reasonable here for perhaps SumerLive alone. But your mileage will vary.

Independent Sites

I’ve covered the more “major” sites with larger staffs, history, and presumably revenue. But that doesn’t mean they are the best sites. I want to highlight some, for lack of a better word, independent sites as I am calling them.

StatRankings

This site has been around for maybe 1-2 years and I’ve had it in my bookmarks folder. When I started putting this post together, I asked for recommendations on sites I might not know of to cover, and I had a few people mention this one. They cover a few different sports, but we are going to stick to football here obviously.

They’ve got plenty of common stats for positions, but also too more advanced stats (which you’ll know require a higher tier of access than the free sub when you see a + sign).

It’s definitely a bit more fantasy and betting focused, but there are a good bit of useful pages and details for the analyst side.

Speaking with someone from the site, where they think they stand out isn’t just trying to have more stats than anyone but in AI integration.

Pretty much every site is trying to build their own AI model or chatbot. While these NFL data sites competition isn’t ChatGPT or Grok or Claude, they are in a way trying to create a version of it. StatRankings instead integrates with ChatGPT, Claude, and Grok and uses their models.

It’s their view that this is the more efficient way to do it, particularly from the website side where you are spending your own resources and server space to build your model. In this case, users can bring the models they already know, with the rate plans and access they already have. In theory too this could lower cost of subscription to sites, since server costs, tokens, etc. aren’t passed off to the user. They are set up to really quickly and cleanly query and pull the user requests and also search engine results.

Like most of the sites, there are some new features due out soon, so expect improvements.

Dynatyze

Another site that covers multiple sports, Dynatyze has some nice tools on the football side, which we are only focusing on here.

It’s really a fun site with some cool features, including things like custom scatterplots (with K means clusters!)

Detailed charting data for multiple positions

I also like this team room page

And this team comparison

Coach pages

I like this player aging page too

And some player contract vs value metrics too

There is too breakdowns for every single game with a ton of details and a nice touch of a card export with some highlights.

Stick to the Model

This site had a cool mock draft simulator and offseason modeler, and has since grown a bit. I think the coolest feature on the site is the Be a GM tool that is in a way Madden Franchise mode

And now they have it for college and NIL too

Ray Carp

Just some really cool and well designed visuals and tools

An on/off tool

And maybe my favorite feature on here in a rage bait pass interference leaderboard

Box Score Scout

Powered by nflfastr/nflverse, team and player level pre-scraped data visualized well

The NFL According to Q

Josh Queipo, who does great contract analysis, has created a nice contract tool called Contract Desk that will give you data and projected deals for mostly any NFL player.

The nice thing about it too is that not only does it give you a projection, but it allows you to include trade leverage and see both sides of the offer, the player and the team as a starting point.

Conclusions

We really have taken a huge leap in NFL data sites over the past year and I am sure I missed something. If there is a site you think I missed, please reach out to me on Twitter (@shauncore) letting me know and I’ll include it in the thread of this Substack. Again, trying to keep it analytics focused, not fantasy or gambling.

Thanks for reading, thanks to all the sites that put in a lot of effort to make the consumer side in a much better shape than it has ever been. Keep going.

If you don't recall (and it’s pre-COVID so I don’t blame you), in 2019 the baseball fandomsphere noticed that home runs were up…way up. On pace to break the single season record by ~500 heading into the all star break. And as it turned out, yes, the league was using a different baseball with less drag. In 2022, MLB commissioner Rob Manfred explained that they were using a new process and more consistent baseballs (which also turned out to be perhaps not true as there were three different balls used that year). The MLB has mandated humidors in all 30 parks but that doesn’t change what happens when the balls move from storage to the playing field, where park atmosphere plays a large part in just how far those balls carry once they make contact with the bat. Meanwhile there is also ongoing independent review and study to prevent future one-off anomalies in production.

The initial sign of a juiced ball in 2019 was actually tipped of by what was going on in the minors, AAA specifically, seeing home runs and slugging percentage spike drastically. With home runs up ~35% over 2018 but no real change in BABIP (which excludes home runs), it was clear the ball was carrying more.

And well, guess what is happening in the minor leagues this year again? Home runs started spiking again.

So has the pattern returned and the major leagues seen a similar random spike in home runs? Yes, it appears something is going on again.

The MLB is by far the best league provider of consumer facing statistics, so much to the point that they even provide us with drag coefficients of the baseball by date.

(science note: drag coefficient measures a baseball’s air resistance as it flies through the atmosphere. It dictates how much a ball slows down due to collisions with air molecules. A lower drag coefficient means the ball experiences less air resistance and will carry much farther off the bat. Every ~0.01 drop in coefficient will produce balls that carry ~5 feet further when hit at 100 MPH)

When drag is lower, batted balls at lower exit velocities carry further than usual, an increase we’ve seen so far this year at pretty unprecedented rates:

And as shown in the drag by date graphic, the coefficient had been climbing in recent years and even was touching near all time highs at the start of the year before just falling off a cliff in May and June.

The xSLG on home runs in June is the lowest it’s ever been in the Statcast era.

The percentage of home runs hit in June that were at or below 100 MPH exit velocity? Also a new high, by quite some measure.

But what we haven’t seen is a spike in exit velocity on home runs (in fact it’s down vs prior years and more in line with 2019 levels).

The same goes for hard hit% and barrel/PA% on home runs, the latter of which is WAY down despite the increase in carry.

There are two main things that impact how far the baseball (or any sports ball that gets hit by something) carry: drag coefficient and Coefficient of Restitution (COR). We’ve covered drag already, but COR is just functionally how “bouncy” the ball is, or technically how much energy is lost on impact. The less energy lost, the more bouncy the ball, the faster it goes.

Since we haven’t seen any increase in how fast balls are moving when hit (in fact we’ve seen a decrease), it doesn’t seem like the balls are more bouncy. COR variations was the root cause of the juiced balls in 2000 and 2019 (though MLB says 2019 was due to drag decrease, however Dr Meredith Wills disputes that and instead was an increase in COR from the center of the ball - literally the core (COR) - being wound tighter).

There has been attempted surface level explanations on the cause of the increase in carry here early-Summer. As I mentioned earlier, the atmosphere has an impact, and an environment with thinner, dry air would lead to further carry than not. So maybe it is just the change in weather? Well no, we’ve been comparing Junes this entire time and the MLB data on drag is adjusted for that. What about a change in elevation and atmosphere? Notably we’ve had a half dozen games Las Vegas, where the air is thinner and drier and the city elevation is about 3x that of the average MLB park (and 5x if you exclude Coors from the data)? No again, the MLB data accounts for that, and such a small number of games wouldn’t have such a higher overall impact. Besides, the drag data had already started to decline prior to those games (June 8th-14th).

What about something like El Nino? This upcoming cycle is expected to be pretty intense. True, but we really haven’t seen that ramp up a ton yet and El Nino is more strong in impact in the winter than summer. And unless we see some change in batted ball distance dispersion between northern and southern stadiums (which I couldn’t find myself through some data querying), it wouldn’t be the cause.

We’ll see how the summer plays out, if this was just a “bad batch” of balls or if this is the baseball we’ll see for all of 2026. We’ve even got MLB data employees investigating and finding similar conclusions on an outside the norm jump in production in June.

There was a lot of talk over the weekend and Monday over the consensus board and reaches and steals. I’m not going to rehash all of the tweets I saw but the general sides are (and I categorize them very broadly):

Pro-consensus board: the consensus board has historically done a good job, as well as NFL teams even, on identifying success outcomes and reaches and steals. There is value in using a consensus board as a data input, particularly in the early rounds. The NFL has generally done better in the later rounds because draft analysts don’t typically scout and talk to 600 players and their coaches, friends, and family in a year like teams do.

(image via Timo Riske at PFF)

“Anti”-consensus board: Teams don’t use a consensus board since they have their own evaluations on players, and specifically look for team and scheme fits rather than general rankings. Teams also have more information than analyst both from a data standpoint as well as being more informed on injury and character red flags.

Anyways, that’s the summation, and I see value in both. Teams certainly have the information advantage. But, and I’m going to switch a bit to my day job vernacular here, they are seeking more active risk.

Skip this section if you don’t enjoy investment and finance talk, but you can compare the consensus board to being like a passive ETF and a team board like an active mutual fund. They do not have the same goal!

A passive ETF is specifically looking for market risk (systematic risk), assuming that the market is efficient, a lot (or all) is priced in, and picking winners/losers is generally not a good idea. Your tracking error (how far you are from the benchmark) should ideally be 0%.

An active mutual fund is seeking active risk (through idiosyncratic risk), where they are going to own maybe 50-60 stocks that they think are winners (could even short expected losers beyond just owning them at 0% vs the benchmark weight), taking on active risk. Your tracking error should be anything BUT 0%. And in fact, if it is close to 0%, you get called a bad name.

NFL teams are trying to pick winners and avoid losers. Consensus boards are trying to take a wisdom of the crowds approach (efficient-market hypothesis). These are not the same goal and they carry different outcomes. And as we have seen historically, active managers have underperformed on the whole vs their benchmarks.

Now to be clear, I don’t think teams should just take the next player on the consensus board blindly. The consensus board should be more used like an anchor to approach and a throttle a bit on overconfidence. If you are a GM and you have the guy that is ranked 99th on the consensus board as a first round pick…you may want to just ask yourself what all these other people are missing. It doesn’t mean you are incorrect or others are, but just that you are probably seeking out a lot of active risk with your grade. And crucially, are you being rewarded for that active risk? Likely not if you take that player at 33rd overall. And while you can still target that player later (which using the consensus board as a general guide is good!), you have to ask on if you want to allow some other team to potentially take on that active risk in the interim between 33rd overall and your next pick.

Okay, now that those arguments are out of the way, the purpose of this piece today is to argue further about steals and reaches. Continuing with my usage the past few weeks with NFL NGS data, it got me thinking (oh boy) on if we can create a sorta consensus board using it to define steals and reaches. Obviously, this isn’t a consensus board since it’s only using one data point. Instead, it’s closer to an expected pick board in a similar vein as the consensus board.

Subscribe now

Gory math details

The goal here is to build a machine learning model capable of predicting an NFL prospect’s draft position strictly based on their physical and college production traits, using historical NGS data. By comparing the model’s “expected pick” to the “actual pick,” we could mathematically quantify which players may have been steals and reaches.

Phase 1: Building the core engine

• The Model: We used a Random Forest Regressor.

• The Logic: A Random Forest is an ensemble machine learning algorithm. It builds hundreds of “decision trees” that look at different random subsets of players and traits, asks a series of mathematical yes/no questions (e.g., Is NGS > 85? Is Position = DE?), and averages all the trees together to output a predicted draft pick.

• The Math (Information Theory): The model determines which traits matter most using Variance Reduction. It mathematically determined that NGS overall score reduced the most “chaos” in the dataset, giving it a 78% feature importance followed by athleticism (8%), production (5%) and size (3%).

• The Result: The baseline model was fairly accurate (explaining 68% of draft variance), but we do have a big blind spot: It only understood absolute trait values, completely missing the human element of the draft. The model can’t incorporate on it’s own medical or character red flags, and I don’t have an large dataset on historical players with those concerns to add as flags for the model.

Phase 2: Feature engineering & the “QB tax”

• The Problem: Quarterbacks with “very good but not elite” traits (like Fernando Mendoza and his 85 overall NGS score) were going 1st overall, but the baseline model expected them to go 11th. The model was punishing them because their raw athletic traits weren’t as freakish as defensive ends.

• The Logic: NFL GMs don’t draft absolute numbers and ignore premium positions surplus value. They draft based on relative scarcity. A team picking #1 overall may reach for a QB simply because he is the best QB available that year and even though he may not go 1.1 every year when there are more/better QBs available, it doesn’t matter if a QB-needy team is picking at the top.

• The Fix: We performed feature engineering. The model created new variables that told the model where a player ranked within their specific draft class. We also added an explicit logic gate (is the player one of the top QBs in the class) to teach the model about the “QB tax.”

• The Result: The model adjusted historic #1 overall QBs (like Mendoza, Joe Burrow, and Baker Mayfield) from the mid-first round up to the top 5 expected picks.

Phase 3: The zero-sum constraint (intra-class ranking)

• The Problem: The model output raw decimal valuations (for instance an expected pick 4.46 or 10.45). But we know the NFL draft is a closed, discrete system. There can be no pick 4.46 or 5.58. Even if the talent gap between the 3rd and 4th best player is massive, someone still has to be drafted 4th.

• The Logic: We needed to convert absolute talent valuations into a realistic draft order.

• The Fix: We applied a discrete intra-class rankings, grouping the players strictly by their Draft Year and ranked the model’s raw decimal outputs sequentially from 1 to N (method = first).

• The Result: The model output is more clean, discrete draft spots (1, 2, 3, 4...) relative only to the players available in that specific year. If Player A is 4.46 and the next spot Player B is 10.45, we treat Player A as being 4th and Player B as 5th (assuming spots 1-2-3 are already taken).

Phase 4: The “true value” integration (PFF pick data + variance)

• The Problem: The standard method of evaluating a draft “steal” relies on counting picks (volume). But the draft board is an exponential curve, meaning we you want to also include franchise impact. A 15-pick slide in the first round can be franchise-altering if you get the right player; a 15-pick slide in the seventh round is generally meaningless. Using just a pure pick difference baseline model treats both equally.

• The Logic: Let’s abandon raw pick numbers. Instead, we need to convert both the “expected pick” and the “actual pick” into a standardized currency to calculate exactly how much free draft capital a GM manufactured or lost.

• The Math: To do that, we integrated PFF Draft Value Chart, which itself is an exponential curve that assigns a specific decimal value to every single draft pick. We then can calculate the value gained (expected pick value minus actual pick value). Then we can use historical value added/lost and normalize for a 1 standard deviation (which ended up being +/-0.08 PFF points) to determine our criteria logic.

• The Fix: The tiering logic uses the standard deviation of value instead of volume. Instead of grouping players by +/-32 picks (a round), we categorized them strictly by PFF value impact: less than 0.5 standard deviation is the “expected range,” 1 to 1.5 is a reach/stel, and beyond 1.5 is considered a “massive” steal or reach.

• The Result: We attempted to eliminate the “volume noise” of the late rounds. For instance, two players in the 2026 draft both slid exactly 11 picks: Dillon Wade slid 11 picks in the 7th round, generating +0.001 value as he ended up a UDFA (pick 258) and the model placed him in the “expected range.” Alternatively, Makai Lemon slid 11 picks in the 1st round, generating +0.35 value (a 2nd-round equivalent), which puts him into the “massive steal” tier as WRs with his NGS data generally go in the top 10 instead of 20th. Using PFF pick values, we can better describe the difference between the 10th and 20th pick in value vs the 247th and 258th pick.

Note: On the images below, you’ll see xPick and xPick Adj. xPick is the raw output, which is why it contains decimals. xPick Adj is the correction for the draft order described earlier. When I say xPick from now on, I am referring to the xPick Adj value rather than the raw xPick.

So who are our steals and reaches? First let’s look through the historical lens. I’m going to look at just the top 170 picks. There comes a point in the draft generally starting in the 3rd or 4th round where teams start to take more fliers on elite traits. The top 170 is ~5th round and I could argue maybe that isn’t strict enough. Also for reaches, we are going to exclude QBs since we know they get pushed up in the draft.

All-time biggest steals

Pretty good list of names! Especially right at the top with DeAndre Hopkins, who via NGS data said should have gone ~8th overall instead of 27th (and to be clear, the model had no idea who DeAndre Hopkins was other than his prospect data so we aren’t cheating here in hindsight). To read the sheet (I’ll give a link to the entire database at the end):

• Hopkins xPick value was 1.042 points (8th overall)

• Hopkins actual pick value was 0.5575 (27th overall)

• His value gained of 0.4849 is the difference between the two of 1.042 - 0.5575

We can then convert his to basically a draft pick uses the same chart, where 0.4849 in value is worth ~34th overall pick. To frame this, we can think of it as a few scenarios for the Texans in 2013:

Scenario A: The Texans draft Hopkins at 8th overall

• They spend the 8th overall pick (1.0424).

• They receive a player worth the 8th overall pick.

• Net Value Gained: 0.00. (a perfectly fair and expected transaction).

Scenario B: Instead of turning the card in, Rick Smith calls the team sitting at pick 27 and offers to trade the 8th overall pick in return.

What is the difference between pick 8 and pick 27? 0.4849 PFF points or the 34th pick. So, to make the trade mathematically fair, the other team gives Houston Pick 27 and the 34th overall pick.

Obviously, the Texans didn’t actually make this trade. They just took Hopkins at 27th overall. But mathematically, the result is as if they did acquire that 34th pick in value by getting Hopkins at 27th instead of 8th.

Of course, this is based on NGS data, so gigantic grain of Morton’s Fine Pink Himalayan Salt here. But we are for fun just treating the NGS model as truth for the xPick. It’s easy to call this a win in hindsight.

As far as the historical reaches:

Here is where using the PFF pick values really come into play and actually hurts the model a bit. Sauce Gardner is on this list! It’s not because the xPick model doesn’t like him (it thinks he’s worth the 10th overall pick in his class), it’s just that it is a pretty steep risk to take him 4th overall instead of 10th. This showcases the steepness of the curve, where 10th vs 4th is worth a lot more than 30th to 24th.

Subscribe now

Okay on to our 2026 biggest steals and reaches. I want to make a quick note here that not every player selected had NGS data. For instance, the 49ers took Kaelon Black at 90th overall, but as you can see below, we don’t have anything for him

So this is a clear gap in the overall scoring, but we can’t use NGS xPick data on players who we don’t have any NGS data.

We already saw Makai Lemon in the historical chart, which goes to show how much the xPick model thinks of him. xPick has him as 9th overall (0.9997 points) and the Eagles got him at 20th (0.6489 points), a gain of 0.3508 points, worth ~52nd overall pick. Not quite Hopkins level, but still really good value on the xPick model. Are we surprised the Eagles are first here?

In my NGS model to create a big board, we already saw how much NGS likes Jake Golday, and that is seen again here in the xPick model where linebackers with his NGS features are taken ~20th overall, giving the Vikings ~62nd pick in value.

A pair of Cleveland and Baltimore players round out the top four, both teams who are analytically friendly and cognizant of the consensus boards and value.

On the reaches side, Tennessee’s pick of Carnell Tate comes in as the biggest reach. At first glance, that seems really weird! He was by many the best receiver in this class. However again as covered in my big board post, Tate had really weak historical NGS data for such a highly ranked WR prospect. He was taken 4th, but his xPick says he should have been ~15th and that Jordyn Tyson and Lemon should have gone before him.

I kept Ty Simpson on the chart even though he’s a QB reach because he matches the narrative too here, where the Rams shocked everyone by him going 13th overall. Even with the QB tax, the NGS data said he should have gone ~47th overall. Of all the QBs taken in the first round in the NGS database, only Brandon Weeden and Dwayne Haskins have a lower NGS overall score than Simpson. Around him are other massive reaches in Josh Freeman, Josh Rosen, Mark Sanchez, and Jared Goff. Joe Flacco is there too, and while in hindsight not a reach when you look at his long career, he seems more like an outlier (and maybe Simpson will be too!)

Just for quick fun, let’s see the first round QBs the xPick model thought were steals:

The model identified Patrick Mahomes, Lamar Jackson, and Justin Herbert as steals?We love it folks. The model thought Justin Fields, Mac Jones, Teddy Bridgewater, and Tua as steals? I pretend I do not see it.

But really, QB evaluation is tough and NGS scores are not great as a whole for QBs, like pretty much any draft eval metric or even pure film grades.

As far a team level goes for 2026, we can break it down a few ways. First is just simply cumulative value added/lost.

Now to be clear, I don’t think the Chiefs had the worst draft. But NGS scoring did not like R Mason Thomas or Peter Woods respective to where they were picked.

And it’s probably being too harsh on Mansoor Delane (though the model doesn’t know the Chiefs traded up for him).

The Eagles come in first place in overall cumulative gain, but they don’t get off completely unscathed. The xPick model did see Markel Bell as a massive reach, but that was offset by nice pickups in Lemon and Eli Stowers.

Next are the Panthers, who didn’t make a single reach (side note: they have a very smart front office and data team)

And the Packers, who didn’t have many picks but added good value

Let’s compare the two New York teams in the Jets and Giants, both who had a good bit of draft capital and first rounders:

Colton Hood was seen as the best pick, but the Jets had all the next best values besides Arvell Reese.

The Jaguars and 49ers caught a lot of flack in the draft for going way off consensus, and the xPick board agrees with the 49ers, but doesn’t hate the Jaguars picks. The pick of Emmanuel Pregnon alone added enough value to offset the two poor values of Wesley Williams and Nate Boerkircher for the Jags.

The 49ers made more than the five picks listed above, but as mentioned earlier, some of these prospects didn’t have NGS data. For the 2026 class, there were 313 players with NGS data and the 49ers somehow ended up with two players without it.

The Cardinals are interesting as they had a lot of PFF pick capital and did not cash in. The xPick model had no issue with Jeremiyah Love at 1.3 from a value perspective (though position value is another conversation), as RBs with his NGS data commonly go that early.

But they followed that pick up with two decent reaches. On it’s face, Carson Beck seems like the bigger reach but the data has the Chase Bisonitis pick as being a greater one.

I thought the Browns had a pretty good draft even before running the data, based off in part they were willing to trade back and just pick BPA that fell. The data bears that out too.

We might be again overrating slightly the difference for Spencer Fano at 1.9 instead of 1.6 (the opposite of Delane) but regardless it was a good move by Cleveland. They were pretty interested in just hitting fairway shots for the most part, even with two smaller reaches in Concepcion and Brailsford.

It was a fun draft season and I learned new things and got to use some new tools and technology to help put this all together.

Below is the 2026 xPick model put link.

https://docs.google.com/spreadsheets/d/1qtEf-XitrQ_CAqjs6lmkD99_9WV0xXbz_l5Nf32UqWk/edit?usp=sharing

First we used the data to look at how good the draft class is historically speaking

Then a bonus post that wasn’t quite NGS data but we used written reports from The Athletic’s The Beast on finding red flags for 2026 prospects be it injury or character concerns.

Then we came to the core of the objective, using NGS data to identify who in the 2026 class is the best/worst by mapping it to PFF WAR

Today’s bonus post is going to just be a quick one on a fun tool I built. Since we have all this NGS and combine (or Pro Day) data, we can create a similarity score dashboard based of 10 features for players.

What we did is ask “Mathematically, who is this player’s closest historical comparison?”

To do this fairly, we of course can only compare position to position, size to size, and grade everyone on their own respective curves.

So the similarity scoring completely ignores raw numbers. Instead, it grades every player on a curve exclusively against their own position. We convert every single stat into a percentile-based grade. It doesn’t look at a defensive tackle and see “310 pounds”; it sees “he is heavier than 85% of all defensive tackles.”

We do this across our ten features. The benefit of using the NGS athleticism and size inputs, is they are scaled themselves already too to a players physical size, similar to RAS scores a lot of us already know and love.

Once every player is graded on this 10-point curve, the model treats each player as a dot floating in a 10-dimensional physical space. It measures the physical distance between the 2026 prospect’s dot and every historical player’s dot.

A 100% match means their dots are literally touching (a mathematical clone - these don’t exist in reality).

An 85% match means they are sitting right next to each other.

A 50% match means they play the same position, but their physical and production profiles are completely alien to each other.

I also included sliders on the dashboard, which you are simply telling the engine to stretch or shrink the room. If you turn the arm length slider to 2.0x, you are telling the computer: “punish players twice as hard if their arm length doesn’t match.” Moving the slider weight up functionally leans into that specific feature more.

Gory math version

The engine relies on a dynamically weighted K-Nearest Neighbors (KNN) architecture, utilizing N-dimensional Euclidean distance on standardized features.

Here is the exact mathematical pipeline:

1. Feature Standardization (Positional Z-Scores)

To normalize the variance across features with vastly different scales (e.g., draft pick ranging from 1 to 259, vs. height ranging from 68.0 to 80.0), every feature is standardized into a z-score.

Crucially, the mean (u) and standard deviation (sigma) are calculated strictly within the player’s listed position.

For any feature x, the z-score is calculated as:

This converts every measurement into units of standard deviation. A z-score of 0 means the player is exactly average for their position. A z-score of +2.0 means they are in the ~97th percentile for their position.

2. Weighted 10-Dimensional Euclidean Geometry

When a target player is selected, the algorithm calculates the spatial distance (d) between the target and every historical candidate (c) in the database.

Because we want the user to be able to dictate the importance of certain traits via UI sliders, we apply a weight vector (w) to the squared differences. The raw distance is calculated using the weighted Euclidean distance formula:

(Where i represents each of the 10 core features)

3. Distance Normalization

Because the user can change the weights (from 0.0x to 3.0x), the maximum possible distance dynamically changes. If a user turns all sliders to 3.0x, the raw distance expands massively, which would artificially tank all similarity scores if we didn’t control for it.

To keep the 0-100 scale stable, we normalize the raw distance by the root of the average applied weight:

If all sliders are left at their default 1.0x setting, the sum Wi = 10, meaning d_raw is simply equal to d_raw

4. The Similarity Score Translation

Finally, we must convert this normalized spatial distance (where closer to 0 is better) into a 0-100 percentage that humans naturally understand.

We apply a linear decay scalar of 15 to the distance. This means that for every 1.0 unit of normalized Euclidean distance (which roughly equates to being 1 standard deviation off across the board), the player loses 15 points from a perfect 100 score.

The max (0, ...) function acts as a floor, ensuring that players who are completely mathematically dissimilar bottom out at 0% rather than dropping into negative percentages.

Subscribe now

Tool breakdown

You’ll select a player on the left hand size either through the search box or the pre-populated players box, which filters as you type

Once you find your player and select them, their model score and data will appear at the top

You’ll find all their NGS data, some combine/Pro Day metrics, their model rank for 2026, and their heuristic tag we covered in the last post.

From there we have our list of 20 closest matches using the above described methodology, as well as that historical players NGS data, their similarity score to the selected 2026 player, and some delta figures showing how far/close that historical player is to the selected 2026 player for each feature.

If you click on a historical player, you are given a radar chart comparing the two selected players using feature z-scores on the plot.

On the left hand side you can adjust the weights. These act as a penalty against a historical player in the similarity scoring finder, or think about it as I said earlier weighing that feature more. The more you move this up, the more it penalizes for distance for your 2026 player for that feature.

There is also the custom compare button, which just let’s you choose any two prospects and see a radar chart for them:

Notice that the players don’t have to be from the same draft class or one from the 2026 draft class. You can just plot the radar charts of any two players and those 10 features. (the weight dials are moot since you are creating your similarity choices).

And that is it. The tool should live on now and forever if you ever want to come back to it.

Link to the similarity dashboard

Happy drafting to all!

Today, we are going to apply what we learned on NGS data and use it to actually try to identify and rank the 2026 NFL draft class.

I could spend a half hour setting everything up here, but let’s just dive in. What I did was take the entire NGS database which has NGS scores for overall score, size, athleticism, and production for essentially every prospect from 2008 to 2026 (regardless of it they participated at the combine or not) and pair it with PFF wins above replacement (WAR) data I was able to source. I then used a random forest model (thanks to the help of Gemini) that looked at each component of NGS and position to try to forecast future WAR. The model was blind to the player’s WAR, so it couldn’t “cheat” and put the known good players at the top of the model board.

I’m going to stop for an important caveat here. Due to my own fault (not saving everything correctly when I had it), I only have PFF WAR data from 2018-2023 and it’s generally for players who had 500+ snaps in that timeframe. Yes, that means there is some survivorship bias in the data, but I don’t think 500+ career snaps is a particularly high bar to jump over. It does mean that some of the players who were like 7th rounders and just didn’t get a chance to play 500+ because they were bad (thus they would correctly weigh down scoring for others) or anyone who was doing well but injured before they reached their 500th career snap, are all missing. Also since the data does not have WAR for 2024-2025 and obviously pre-2018, it won’t fully capture players who maybe broke out in 2024 or anyone who was great in 2017 but didn’t play anymore after that. So I’m the first (and last because why would there be anyone else?) to admit it’s an imperfect system here and I wish we had everything for everyone. But I don’t think we are missing THAT much. We have five years of WAR data for every player who reached 500 snaps total in 2018-2023.

The full outlay is below, with some wonky details and gory math if you’d like.

Part 1: The Quantitative Factor Model (The Gradient)

The core predictive engine is a supervised machine learning algorithm built to project WAR.

1. The Algorithm: Random Forest Regressor

We selected a Random Forest Regressor rather than a standard Multiple Linear Regression. Football traits do not scale linearly. For example, adding 10 pounds to a 240-pound Edge Rusher adds significant WAR value, but adding 10 pounds to a 190-pound Cornerback might destroy their agility and tank their WAR. A Random Forest uses an ensemble of decision trees to capture these complex, non-linear relationships and conditional variable interactions.

2. The Training Data & Target Variable

• Target Variable (y): Actual NFL WAR

• Training Set: We isolated the draft classes from 2016 through 2022. This specific window provides a modern sample of athletes while ensuring players have been in the league long enough to accrue a stable, verifiable NFL WAR metric.

3. The Input Features (x)

The model ingested five structured, continuous arrays for every player:

1. expected_pick: for players in the 2026 draft we used their consensus big board rankings as of last week, for everyone else we used their actual draft pick, with UDFAs all marked as the 258th pick

2. production_score: how a player performed in college per NGS scoring

3. athleticism_score: position adjusted athleticism data from NGS

4. size_score: physical dimensions (height, weight, arm length, and hand size).

5. ngs_score: The overall Next Gen Stats score

4. Hyperparameter Tuning & Regularization

To prevent the model from simply memorizing the training data (overfitting), we instituted strict constraints:

• n_estimators = 200: The model builds 200 distinct decision trees on random subsets of the data and averages their outputs, drastically reducing variance.

• min_samples_leaf = 4: This is our primary structural constraint. We forced every terminal “leaf” of a decision tree to contain at least 4 historical players. This mathematically prevents the model from looking at a single historical outlier (like Tom Brady) and building a brittle rule around them. This also solves for players with little comparable data such as inn the 2016-2022 training set, only two edge rushers have gone first overall (Travon Walker and Myles Garrett).

5. Out-of-Sample Validation (The “Honesty Test”)

To validate predictive accuracy, we utilized 5-Fold Cross-Validation (cv=5). The 2016-2022 dataset was split into five distinct chunks. The model was trained on four chunks and forced to predict the fifth chunk completely blind. This guarantees that the WAR projections driving the board are based on true out-of-sample predictive power, not data dredging.

6. Output Normalization (The Z-Score)

Raw WAR is heavily skewed by positional value (a replacement-level QB generates more WAR than an All-Pro Guard). We applied a standard scaler to the model’s output to convert Raw Expected WAR into the Pos_Agnostic_Score. This translates the projection into a Z-Score (Standard Deviations from the Positional Mean), creating a universally comparable gradient across all positions.

Part 2: The Risk Management Overlay (The Cliffs)

Machine learning models are highly susceptible to “Trait Seduction.” If a player has a 99-percentile Athleticism Score but a catastrophic 50-percentile Production Score, the Random Forest might average them out and output a “safe” +0.50 Z-Score.

To prevent the algorithm from smoothing over fatal, non-linear flaws, we built a deterministic, rules-based logic gate system: The 14 Heuristic Tiers. These are hard mathematical thresholds (cliffs) based on historical dealbreakers.

Technical Threshold Examples:

• The Tape Cliff (Workout Warrior Trap): * Logic: IF Expected_Pick <= 50 AND production_score < 70.0 -> Assign Tier 14.

  • Rationale: Elite physical traits cannot compensate for a sub-70 production score in the top 50 picks. The bust rate hits critical mass.

• The Physics Cliff (Undersized DT Trap):

  • Logic: IF Listed Pos == 'DT' AND Expected_Pick <= 100 AND size_score < 65.0 -> Assign Tier 12.

  • Exception Override: UNLESS ngs_score >= 95.0.

  • Rationale: A lack of baseline mass is a fatal flaw for interior defensive linemen unless they possess literal 95th-percentile, Aaron Donald-level twitch to transcend physics.

• The Volume Trap (Fools Gold RB):

  • Logic: IF Listed Pos == 'RB' AND Expected_Pick > 100 AND ngs_score >= 85.0 -> Assign Tier 10.

  • Rationale: RB WAR is heavily dependent on carry volume. Day 3 running backs, regardless of their elite combine testing (>85 NGS), rarely receive the requisite NFL touches to generate positive WAR.

Part 3: The Master Sort (Synthesizing the Board)

The final War Room Big Board is generated by passing the data through a strict, dual-key sorting algorithm. This synthesis marries the risk-management parameters of Part 2 with the quantitative probability curve of Part 1.

The Dual-Key Sort:

1. Primary Key (Ascending): Tiers (1 through 14).

   • The algorithm first groups all players strictly by their heuristic bucket. This is an absolute constraint. It physically guarantees that a Blue-Chip Lock (Tier 2) will always be drafted before a High-Floor Value (Tier 4), and ensures that all TRAP players (Tiers 10-14) are quarantined at the absolute bottom of the board.

2. Secondary Key (Descending): Pos_Agnostic_Score (The Z-Score).

   • Once the players are locked inside their respective risk-adjusted tiers, the algorithm sorts them internally by their expected WAR Z-Score.

The Output Variables: Executing this dual-sort generates two final index columns:

• Draft_Year_Blind_Rank: An intraclass index used to identify the absolute best player available on the board for a specific draft year (e.g., the 2026 class).

• Global_Blind_Rank: An interclass historical index used to evaluate how a current prospect stacks up against the entire 2016-2026 database on an identical quantitative scale.

And as always we want to be clear of the flaws and drawbacks in the methodology that are worth considering:

The “Expected Pick” Echo Chamber

To train the Random Forest, we used expected_pick as one of the five core inputs. The logic is sound: NFL scouts are generally smart, and draft capital is historically the single highest predictor of success.

• Drawback: We have baked “Consensus Bias” directly into the math. If the entire media and league echo chamber is artificially inflating a Quarterback because of his arm strength, our algorithm will see that high Expected Pick and anchor to it, artificially boosting his Z-Score. It makes it harder for the model to go truly contrarian on high-profile players.

The Noise of NFL WAR (The Target Variable)

Our model is built entirely to predict one number: Wins Above Replacement. In baseball (where WAR was invented), the game is a series of isolated 1-on-1 events. Football is an infinitely complex 11-on-11 ecosystem.

• Drawback: NFL WAR is incredibly noisy. An elite WR drafted by a team with a terrible QB and a run-heavy scheme will generate very low WAR. An average CB playing opposite an All-Pro edge rusher might generate inflated WAR. The model is training on a target variable that is heavily influenced by coaching, scheme, and supporting cast. These are things that a college prospect cannot control.

The Margin of Error on the “Cliffs”

We built the 14-Tier heuristic system to act as strict risk-management cliffs. For example, the workout warrior trap requires a college production score of 70.0.

• Drawback: Hard cutoffs are mathematically ruthless to players sitting directly on the margin. A player with a 69.9 production score gets banished to Tier 14 (the trap tier) and falls hundreds of spots down the board. A player with a 70.1 production score safely lands in Tier 5 and remains draftable. In reality, there is zero football difference between a 69.9 and a 70.1, but the algorithm treats them as completely different assets.

• So you may ask “why not use a hill rather than a cliff?” Think of all of this as a risk management tool. If we used hills over cliffs, we’d be allowing some negotiation in scoring.

  • A player with a 99 athleticism score but a 51 production score would be graded higher because you’d be striking a balance between the two. But that isn’t what we want here. A 51 production score is a non-starter. A corner who runs a 4.3 forty doesn’t matter if he hasn’t show he can actually cover receivers.

  • Sometimes failure is a gentle slope, but in football, it’s a quick drop-off. Historical data has shown that there is a real difference between 280 and 290 pounds at defensive tackle. We are facing the physical laws of the sport.

  • The board is setup to be a decision making solution. The tier separations defines clear hierarchies. It doesn’t really help us if one player has a 43% chance of 5 WAR and the other has a 45%. We want to create clearer, definable rankings when we can.

Context-Agnostic Production

The model ingests production_score as a pure number.

Drawback: Math cannot watch the tape and explain a players production fully. This is in part why we include draft pick information as a variable to help out a bit here in bridging the tape and the data.

Now let’s dive into that 14-tier heuristic system we set up, since it goes into the scoring methodology as well as gives us some explanation on to why a prospect is ranked to where they are at.

Tier 1 Generational Alien

• Factors: Z-Score > 5.0 AND Expected Pick <= 15 AND Production Score >= 75.

• Intent: Identifies prospects who mathematically break the positional scale. The intent is to isolate players possessing an extraordinarily rare blend of size, speed, collegiate dominance, and usage of draft capital. Generational alien types do not commonly last more than 5 picks. There are only 12 players who have been in this tier since 2008 and the latest in the draft one was picked was Anthony Barr at 9th overall.

• Highest ranked recent model examples: Derrick Brown, Quentin Nelson, Myles Garrett, Saquon Barkley.

Tier 2 Blue-Chip Lock

• Factors: NGS Score >= 85 AND Expected Pick <= 50 AND Production Score >= 75.

• Intent: Highlights the safest foundational pieces in the draft. The intent is to flag prospects who pass every strict threshold for both athletic traits and functional football tape.

• Highest ranked recent model examples: Derwin James, Devin White, Travis Hunter, Minkah Fitzpatrick, Marvin Harrison Jr, Bijan Robinson

Tier 3 Elite Edge Rusher

• Factors: Position = Defensive End AND NGS Score >= 85 AND Expected Pick <= 100.

• Intent: Acts as a positional premium override. Raw athletic traits translate more directly to edge rusher success than almost any other position, validating an elevated tier purely for elite athletes on the defensive edge. Spoiler: this tier is blank, as no edge rusher has qualified.

• This is one I want to go back and revisit after the draft. Working with the model to create tiers, all these passed the initial concept but when the data was run, this tier was left blank. The concept is sound for this tier to exist, as teams value edge rushers very highly in the draft, and those players with great NGS scores and high expected pick values are swooped up quickly on day one. However with our scoring rules, tiers 1 and 2 vacuum up any players who would qualify for this tier.

• Here are our tier 4 edge rushers who went in the top 15 (or in 2026 are expected to). As mentioned in the drawbacks section, draft spot is anchoring things a bit here. In hindsight, Clelin Ferrell should not have gone 4th overall, and he was 19th on the consensus board. But we don’t want to use hindsight here because we don’t know the outcome of the 2026 class. Right now Rueben Bain and Arvell Reese are in tier 4 (which is a good tier of players mind you). We don’t want to move them to tier 3 when it’s possible they go close to 4th overall like Ferrell and end up a bust too.

  

Tier 4. High-Floor Value (Trust the Tape)

• Factors: NGS Score between 65-84 AND Expected Pick <= 50 AND Production Score >= 78.

• Intent: Isolates highly productive college players who lack top-end athleticism. The intent is to secure safe, high-floor assets who demonstrated sustained collegiate dominance against top competition.

• Highest ranked recent model examples: Javonte Williams, Devon Witherspoon, Edgerrin Cooper, Elijah Moore, Vita Vea, Jalen Carter

Tier 5: Draftable Starter

• Factors: Expected Pick <= 100.

• Intent: Classifies baseline, competent NFL prospects who avoided all major trap thresholds but did not trigger the parameters required for elite categorization.

• Highest ranked recent model examples: Anthony Miller, O’Cyrus Torrence, John Michael Schmitz, Andy Isabella, Trey McBride, J.K. Dobbins

Tier 6: Late-Round Steal (Production Darling)

• Factors: Expected Pick > 100 AND Production Score >= 80 AND Position = Premium (QB, SAF, IOL, LB, WR, TE).

• Intent: Targets the safest gambles in the late rounds. The intent is to draft elite collegiate tape at positions that heavily rely on mental processing and technique rather than pure physics.

• Highest ranked recent model examples: Antonio Johnson, Amani Hooker, Troy Franklin, Armani Watts, Shedeur Sanders, Mason McCormick, C.J. Gardner-Johnson

Tier 7: Lottery Ticket (Elite Traits)

• Factors: Expected Pick > 100 AND Athleticism Score >= 85.

• Intent: Identifies raw developmental projects. The intent is to draft strictly for the highest possible physical ceiling, accepting the lack of college production for the chance at securing unteachable physical traits.

• Highest ranked recent model examples: Jakorian Bennett, Channing Tindall, Leo Chenal, Chimere Dike, Theo Johnson, Dont’e Thornton Jr., Julian Love, Maxx Crosby, Baron Browning

Tier 8: Value Flier (Production Slipper)

• Factors: Expected Pick > 100 AND Production Score >= 75.

• Intent: Captures prospects who produced well in college but fell in the draft due to poor athletic testing or other issues. The intent is to acquire high-floor depth pieces at a discount.

• Highest ranked recent model examples: Dalton Keene, Alex Highsmith, Blake Freeland, Kelee Ringo, Yodny Cajuste, Dawand Jones, Cedric Gray

Tier 9: Day 3 / UDFA

• Factors: All remaining players with Expected Pick > 100.

• Intent: Groups the baseline late-round players who lack both elite traits and elite production.

• Highest ranked recent model examples: Tai Felton, Chris Herndon, Josey Jewell, Adam Trautman, Jeremy Ruckert, Mark Walton, Cameron Latu

The “Do Not Draft” Zone (The Traps)

Tier 10: High NGS RB (Fools Gold)

• Factors: Position = Running Back AND NGS Score >= 85 AND Expected Pick > 100.

• Intent: Flags fast running backs drafted late. The intent is to avoid prospects whose athletic testing looks appealing but who will mathematically fail to receive the necessary NFL carry volume to generate positive value.

• Highest ranked recent model examples: Cam Skattebo, Israel Abanikanda, Jaylen Wright, Chase Brown, Sean Tucker, Elijah Mitchell, Dylan Sampson

11. 🛑 Specialist

• Factors: Position = K, P, LS, FB.

• Intent: Acts as a strict positional value override. The intent is to prohibit spending premium draft capital on special teams positions or fullbacks.

12. 🚨 TRAP: Undersized DT

• Factors: Position = Defensive Tackle AND Size Score < 65 AND Expected Pick <= 100 (Override: NGS Score > 95).

• Intent: Enforces the laws of physics in the trenches. The intent is to prevent the drafting of interior linemen lacking baseline mass, as failure rates skyrocket against professional offensive guards unless the player possesses 95th-percentile twitch.

• Highest ranked recent model examples: Ed Oliver, Byron Murphy II, Calijah Kancey, Xavier Cooper, Will Sutton, Jurrell Casey

13. 🚨 TRAP: Low-Prod CB/OT

• Factors: Position = Cornerback or Offensive Tackle AND Production Score < 65 AND Expected Pick <= 100.

• Intent: Penalizes poor technique at highly technical positions. The intent is to avoid raw athletes at cornerback and offensive tackle, as NFL speed immediately exposes a lack of fundamental production and technique.

• Highest ranked recent model examples: Upton Stout, Renardo Green, Cordale Flott, Caedan Wallace, Justin Walley, Jarrian Jones, Emery Jones Jr., Wanya Morris

14. 🚨 TRAP: Workout Warrior

• Factors: Expected Pick <= 50 AND Production Score < 70.

• Intent: Flags the single highest bust profile in the sport. The intent is to explicitly forbid drafting players high in the draft strictly based on combine numbers when they failed to reach a functional baseline of production against collegiate competition.

• Highest ranked recent model examples: Travon Walker, Luke Musgrave, Chris Lindstrom, Anthony Richardson, Jalen Reagor, Mykel Williams, Drew Lock, Odafe Oweh, Will McDonald IV

Now to be clear, there are some hits and misses in each tier, and that is what should happen. No tool is perfect, this is not meant to be a crystal ball. Any model that is perfect was clearly overfitted. It’s nice that the model labeled Maxx Crosby as a steal. It’s not nice it said Ed Oliver should have been a day three pick. But we don’t want a model that gets everything right 100% of the time, or it’s not a model. It’s just a describing outcomes rather than projecting.

Think of all of this as a probabilistic risk-engine, like determining what to do in hands of blackjack. A blue chip prospect will be good 80% of the time, but that means 20% of the time he will be a bust. Standing on 20 when the dealer ends up with 21 isn’t bad process, it’s variance.

Also think about these in terms of tiers too. Generally speaking, the tier 1 and 2 guys (aliens and blue chips) don’t often bust because they were bad players, it’s those invisible variables we can’t really account for fully. Devin White went 5th overall and was loved by the data and the film. He is the 16th highest scoring player in the model data. But nobody, the data or film grinders, could account for his injuries, freelancing, and his mental health struggles. It was impossible for anyone to be right.

Tier 7 guys are by definition lottery tickets, so you expect a lot of misses here. But this is why you are taking them outside the top 100, where the cost of missing is low and the risk/reward is generally asymmetric.

Maybe too it will be helpful to look at some successes and failures of the model. Let’s just keep it simple and look at where the model ranked a player in their draft class against where they went. It’s not useful to say “the model really liked Myles Garrett” because everyone and everything did.

Subscribe now

Let’s start with the guys who went outside of the first round but the model said should have been first rounders. Not a perfect hit rate but we’ve identified some players the model classified as blue-chip locks and that ended up being a pretty good categorization:

Some of these guys the model said forget just being a first round pick, they should be top 10 picks. Overall hits include Jaylon Smith, Cooper DeJean, Eric Kendricks, Brian Branch, Breece Hall, Rob Gronkowski, Derrick Henry, Dalvin Cook, Jonathan Taylor, Dallas Goedert, Deebo Samuel, Nick Emmanwori.

Blue-chip locks (ex-QBs) have an average of ~0.8 WAR (for players who have WAR values), which is roughly the WAR between 2018-2023 that Montez Sweat, Dalvin Cook, Eric Kendricks, Jabrill Peppers, Rashawn Slater, and Chase Young all put up.

Now let’s look at some guys where the model said “you took a guy in the top 20 when he was a 2nd rounder at best”: Marlon Humphrey, Chris Lindstrom, A.J. Terrell, Garett Bolles, Odell Beckham Jr., Adoree’ Jackson, Arik Armstead, Taylor Decker, Rashan Gary, Haason Reddick, and Ed Oliver as examples.

Outside of Ed Oliver (who we mentioned earlier the model thought was too small), these players didn’t have great NGS data, averaging an overall NGS score of ~81 and mostly weren’t productive or athletic. Now to be clear, outside Lindstrom and Oliver, the model didn’t say they stink. They still had them as draftable starters, just not top 20 picks overall.

There are three players drafted in the top 5 that the model said were closer to the 300th best prospect in the class than the 1st: Travon Walker, Dion Jordan, Anthony and Richardson. Now to Walker’s credit, he has been better and signed that major deal, so we could call that a miss, but the model said he was clearly not one of the 20 best players in the draft.

Now, on to the 2026 class. Our top 20 prospects:

Jeremiyah Love hold the highest NGS score for any RB in the dataset history, but falls short of being an alien because of his size. The model does punish RBs for their size a bit since it is trained off WAR and WAR needs playing time. Still, he comes in as the best player in the class.

Styles, Mendoza, and Bailey I think everyone can agree make sense up at the top. Kenyon Sadiq looks like a player the model is high on, but for him a lot of it just comes down to positional value rather than skill. Still overall, he’s the 12th best tight end prospect in the model history and those around him are mostly good hits.

The last two blue chippers are Jordyn Tyson and Spencer Fano. Going back to what the data doesn’t know, I know there are reports that several teams think Tyson is the best WR in the draft, but health is a major concern. While I did mess with adding scouting report data (more on that in a future post), I couldn’t quite get it to work as well as I hoped. If we included that, we could tell the model that Tyson is often injured, and it would likely ding him a bit. However, we have to just rely on that being baked into his consensus big board rank of #12.

Some comps for Tyson

A lot of really good players in the top range there. Kadarius Toney yeah, yeah, if we knew he was more focused on his rap career than football, he probably wouldn’t have gone 20th overall (and fwiw the model had him as WR6 in 2021).

Fano might not be the most physically imposing tackle, but highly athletic and productive OTs that are also seen as consensus top picks have generally good outcomes.

Let’s look at our tier 4 players, who the model thinks are still good despite maybe the verbiage of “high-value floor”.

Really these guys are being “knocked” for being productive in college but not jumping off the figurative tape for athleticism. But the model is aware that there is something it isn’t capturing when you combine the production for the player and position with the other data. These guys are all clear first rounders.

For reference, here was the 2023 class of tier 4 guys. Not a 100% hit rate but again we wouldn’t expect that. However, the high-floor moniker is not a negative, as there are clearly talented and successful players below.

The guy on our board that stands out is Jake Golday, a linebacker out of Cincy. On consensus he’s ~52nd (my similarity dashboard was built a few days ago so it’s not 100% current which is why he’s 48th right now), but the model thinks he’s a top 10 player in his class.

Four of the guys above went in the ~first round, with several pretty good players here. Golday finished second in this linebacker class in both production and overall score, which is fair because he was only beat out by Sonny Styles who is clearly the better prospect.

Compare the second tier linebacker class, even if you don’t buy Golday as a top 10 player in the draft (which is fair and why we have the blended risk column to try to merge both the model and the consensus pick to re-anchor our scoring), you can make the case based on the historical data that he’s worth 20-30 higher of a pick than he’s being valued at.

To be clear, even if you fully buy into the model, you shouldn’t take him at 10th overall, because of everything we know about reaches and steals in the NFL draft. If you can get Golday at 45 instead of 10, then you should figure out a way to take him at 45 rather than 10, even if you think he should be 10th, because presumably you can trade back from 10 (or just take someone else) or you can trade up to take him where you’ve de-risked some of the potential to be massively wrong at 50th overall vs 10th.

That tier 4 group has four cornerbacks (Avieon Terrell, Jermod McCoy, Mansoor Delane, and D’Angelo Ponds).

A recent-ish history of corners given that label:

It’s a pretty good group, but these guys do tend to go maybe a little closer to 20th than 10th. All signs point to Delane being the first corner off the board, possibly taken in the top 10. But maybe it makes more sense to group him a little later (McCoy notwithstanding who is certainly due to fall due to his knee issues and the model doesn’t know he may not be able to have a long career).

Onto Tier 5: Draftable Starter, I’m going to skip posting the image because it makes up 69 (nice) players in this draft class. I’ll have a link to the entire board and data at the end here. But here are the biggest differences between the model and their consensus rank (difference column):

All the above are players that are 25+ picks +/- their consensus rank. It’s a lot of guys in the trenches who the model thinks are overrated. Many with low production scores and all with sub-81 NGS overall scores. Good players mind you, but at odds with positions and scores that have done well (per WAR).

But I want to add a quick note here, and I’ll quote former Citibank CEO Chuck Prince: “As long as the music is playing, you’ve got to get up and dance.” It may not be that Keldrick Fault in the average draft year would go ~19th overall, and that probably wouldn’t be the case in a really good draft class at the top. But if you are a team picking 19th overall and defensive end makes sense, you have to dance. You can’t just keep trading back until everyone is at the right spot on your board. Trade backs aren’t frictionless and require two people, to continue the dance metaphor, to tango. At some point, you’ve got to pick a player and you may not love the player at that pick, but you can’t control that. So all of this to say, I don’t blame necessarily someone taking a player maybe a bit higher than they might call for, because you have to dance.

And the inverse is true, there are players with NGS data who historically have lead to better outcomes than their expected pick range implies.

Take Jake Slaughter from Florida for instance:

Players with his general NGS, expected pick, and size data typically go earlier on day two, if not already on day one.

Same for Devin Moore

Deion Burks

And Julian Neal

It’s not that these guys are great or amazing prospects, it’s that players with similar numbers typically are priced a little higher than where they stand on the consensus board.

For instance, here is the radar plot for Julian Neal vs his highest similarity player in Trevon Diggs.

Neal is on par data-wise with Diggs in all but expected pick, which for Neal is his consensus board rank (100th) and Diggs who was taken at 51st.

This again isn’t to say take Neal at 51st, but that his profile is intriguing enough in the NGS and combine data that he might be undervalued if he’s expected to go closer to 100th than 50th.

Subscribe now

Let’s move to tier 6 in late-round steals

Looks like a small group, but since 2008 the model only has assigned that tier to 62 players. While I would say this tier has had some hits (C.J. Gardner-Johnson, Amani Hooker, Harrison Bryant, DeShon Elliott, Isaiah Likely, and Talanoa Hufanga to name a few), I think it’s best used as more of an anchor than something to seek out. By definition, these are guys the model sees are being undervalued but not that it is saying should go a ton earlier. It’s looking at guys it models as being 3rd round+ talents, but expected to go later in the draft.

Tier 7 are players given the heuristic tag of lottery tickets. Player with elite athleticism scores but generally lackluster production in college.

Our example here will be Bryce Lance, who has the 7th highest athleticism score for a WR in NGS history and it wasn’t like he was completely unproductive.

When we look at wide receivers with an athleticism score of >90 and a production score >75, the hit rate is pretty good.

Now let’s be clear, you have to fold the tape into the analysis too, and it’s likely Lance is more like an outlier. Despite posting insane athleticism numbers, he’s ranked ~100th in the consensus board vs all those other guys who were generally considered day 1-2 picks.

Christian Watson seems like the easy comp: both went to North Dakota State, both posted ~99 athleticism scores, and both ~80 productions scores. However, Watson was a little bit bigger all around, younger than Lance was coming out, and Watson had more explosiveness and fluidity to his game and usage.

And this is too why we trained the model on draft pick spot, because without it, it’s possible that Lance and Watson would have the same model score, whereas with the pick expectation, the model has them several tier of difference.

But still, even if you say Lance is only 80% of Watson, Watson went 34th overall (Packers even traded up to get him), whereas Lance is being priced at closer to the end of the 3rd.

Tier 8 is our prospects who were productive in college, but don’t jump off the page in athleticism and the film grade didn’t love them either.

Bud Clark being here is a nice validation of the model concepts, who draws high reviews for being an intelligent and playmaking safety but lacking the standout athletic traits and physical characteristics.

There are a trio of running backs here that the model likes the production of modestly, but remember that WAR (or projected WAR for 2026 prospects) is part of the model and the model knows that in order to accrue WAR, you need snaps, which means you cannot be a small ball-carrier. Those guys are right on the line of the size and productive balance.

Tier 9 is our biggest list, it’s 175 players that are being considered as Day 3/UDFA caliber. Again, you can find the full list at the end link, but here are the top players on the consensus board that are in this tier.

16 players are in the 100-125 range that are in this tier, with the below players being in the wider gap of model rank vs consensus board.

We’ll speed through tier 11, it’s a fullback. Sorry Max Bredeson.

Tiers 12 and 13 are players the model flagged and created tiers for where historically their NGS inputs have had a fatal flaw. For defensive tackles, small size has not worked out (as mentioned earlier, a physics problem). And for corners, production has been important.

Let’s dive into corners really quickly looking at which traits have driven WAR.

80th+ percentile WAR:

Production ~77.3

Athleticism ~74

Size ~75.3

Overall ~77.6

40-60th percentile WAR:

Production ~70.6

Athleticism ~71.9

Size ~71.9

Overall ~72.5

Sub-20th percentile WAR

Production ~68.8

Athleticism ~75

Size 75.5

Overall ~72.1

Subscribe now

And now our final tier, what is deemed as workout warriors. Players who weren’t productive in college, aren’t particularly big (outside of Caleb Banks), and only really two calling cards were good combine/pro days and rank highly on the consensus board.

Now the consensus board rank here matters too, so there is something on the tape that analysts like beyond the stats and physical profile. There are analyst out there that for instance have Max Iheanachor going in the first round. And that explains the disconnect, because Iheanachor is pure projection. He’s newer to football and physically big and athletic.

And it’s the same with Ty Simpson, who might not only be a day one pick but given the QB premium and lack of QBs in the class, a team with multiple firsts could consider taking him with their second in the opening round. This is despite a middling production, athleticism, and size score. But the film overrides the data a bit in analyst rankings because you can see the traits and projection, while he played in a tough conference, and this was just his first year starting.

Two things I’ll leave you with, a snapshot of the top 80

And a link to the entire model for the 2026 class is here

Tomorrow I’ll publish the link to the dashboard with some more information on it, or you can find a link to it right now on my Twitter.

I believe too for the first time, it was released both as the usual PDF version (which I prefer) and also a digital and interactive version. Both look nice!

What Dane does is more than just rank 2,700 prospects or capture details about them. For many of them (a few hundred) he writes detailed prospect capsules on, breaking down their background, history, traits, strength, weaknesses, stats, and summaries. What is neat here for our article is that we can extract some of the nuggets Brugler has mentioned to look for specific things. In this case, we want to look at players who Brugler has mentioned (either explicitly or contextually) as having red flags. This can be either a history of injury, or what is usually called off field issues (legal troubles, attitude issue, lack of effort, etc).

Here is what we asked our model to do:

• Scan The Beast for scouting report text for signals tied to injury/durability and off-field/character concerns

• Do no rely only on exact keywords. Look at context and meaning, so phrases like “can’t stay on the field,” “durability will be questioned,” or “maturity/accountability concerns” are treated as red-flag language even when the wording varies

• Match those passages to the correct player, position, school, and page number.

• Cassify each finding into categories like risk type, severity, and subtype.

• Summarize and normalize the findings into a spreadsheet format for public sharing

If you want the full technical explanation (from the LLM itself):

I ran a retrieval-augmented, schema-constrained information extraction pipeline over a long, semi-structured scouting PDF, using an LLM for passage-level semantic adjudication, risk labeling, and source-grounded normalization.

This was a high-recall candidate retrieval + zero-shot semantic classification + structured extraction workflow. I used lexical triggers and phrase families to surface candidate passages, then used the model to decide whether the surrounding context actually expressed an injury/durability or off-field/character risk, even when the wording was indirect.

So with all that wonky details out of the way, here is the results (including page citations from the guide).

Subscribe now

Full version:

And sorta the slimmed down all-in-one image version.

Here is a link to the full output via Google Sheets.

All report details and credit for the information goes to Dane Brugler and The Athletic. You can subscribe to The Athletic here. Model used was OpenAI GPT 5.4 Pro

I don’t pretend to be an expert on AI, bodily injuries, or NFL prospects (or generally anything other than being annoying on Twitter). But this was an exercise personally to try something a little different, while also trying to be informative and useful to the analytics community (and timely with the draft).

Have a good weekend.

Subscribe now

Using NGS scoring data that goes back to 2008, I’m going to be looking at the 2026 NFL draft class in a few different ways. Today we’ll just be using some pre-built NGS metrics to describe the class and put it in historical context.

As a primer on NGS scores, back in 2019 the NextGen Stats team at NFL went about tackling a draft model to quantify and rank players using four (now three) categories:

1) How athletic is a player based on measurable drills at the NFL Scouting Combine?

2) How productive was a player in college based on on-field performance?

3) How big is a player relative to other players in their position group?

4) How will a player perform in the NFL based on their athleticism, production and size profiles?

I said only three now because the NGS data stopped creating a size score for prospects and only does athleticism, production, and an overall score now. I ended up using some machine learning using height, weight, hand size, and arm length of prior prospects, already knowing their size scores and measurements, to create a size score for prospects who are missing that data (players from 2023-2026).

I also swear at one point the production score was not just college stats, but also incorporated draft rankings to make a sorta tape + stats figure, but I can’t seem to find that literature anymore.

I’ll give a little more of the gory details from the linked page above:

It can be helpful to think of the model as several different position-specific models, tailored specifically to distinguish between the traits needed for success at each position group, with players separated into the following positions: quarterback (QB), running back (RB), wide receiver (WR), tight end (TE), offensive tackle (T), guard (G), center (C), edge (ED), defensive tackle (DT), linebacker (LB), cornerback (CB) and safety (S).

The models use a decision-tree-based algorithm called XGBoost to predict the likelihood that the player will become an NFL starter or Pro Bowler within the first three seasons of his career. The models are trained on historical data from the NFL combine (since 2003) and on-field college statistics (since 2005), with rigorous feature selection techniques applied to each position-specific model. The resulting probabilities are converted into composite scores for each player -- representing athleticism, production, size and final overall score -- driven by the key traits that best predict NFL success.

The cool thing is that they didn’t just run it for the 2019 class, but also gave us all the data for classes back to 2008 and continuously each year. That means we have 5,992 players in the NGS database, including this year’s prospects. That’s a huge sample size of data and that’s even after we’ve removed some of the oddity positions like punters, kickers, fullbacks, and long snappers. Nearly 6,000 players, most of whom we know how their careers went, giving us a look at how well these scores work.

So over the next few articles we are going to do a few things. Firstly today, we are just going to look at the quality of this class vs prior years. Fairly simple. But we’ll also try to use NGS scores and some other data to model prospect outcomes, find which NGS traits lead to success, and try to build a similarity scoring methodology to create prospect historical comparisons. I guess buckle up if this sorta stuff revs your metaphorical engine.

Okay, onto topic #1: quality.

Generally when assessing quality of a draft class (not in hindsight but beforehand) I think the two methods are:

1. IDK let’s just eyeball it and go with our gut?

2. Rank every player and then just look at the positions that are high and low relative to others

The first process…ain’t great. It’s very, very subjective. And while the second process has subjectivity too (if it is just YOUR rankings), if you use some sort of consensus ranking, you can arrive at a generally objective estimate. But this still mainly just stacks positions in order, rather than perhaps ranking the class as a whole or looking beyond just the high level figure of their big board rank.

I think the general vibe is the 2026 draft class isn’t great, at least overall relative to other years. To the point where we’ve seen talking heads say that teams may be valuing 2027 picks rather than focusing on acquiring picks for the 2026 draft (which I think is wrong to do in most cases but whatever).

There are some aspects that I think are both subjectively and objectively true though. Like the QB class this year is thin, and I don’t think there is any argument using tape, stats, or rankings that it isn’t. This isn’t a knock on Fernando Mendoza, who is a fine 1.1 overall candidate as any, even if he isn’t at the NGS level of Robert Griffin III (highest QB prospect score in the database), or the next highest of Jayden Daniels (yes you read this right), Andrew Luck, and Kyler Murray. Mendoza sits with the 17th highest NGS score of any QB prospect, right in the middle between Tua Tagovailoa, Patrick Mahomes, Zach Wilson, Mendoza, Cam Ward, Baker Mayfield, and Lamar Jackson. Baker and Ward both went first overall. Tua likely would have in many classes but his own. Zach Wilson was in that mess of a 2021 QB class. And it’s clear Mahomes and Lamar were not valued properly on big boards relative to NGS. A bit of kudos to NGS who had Mahomes as the #2 QB in the class just barely behind Watson, but a huge gap between Watson/Mahomes and Mitch Trubisky, despite Trubisky going 8-10 picks earlier. Also having Lamar functionally tied with Baker - and Lamar incorrectly went 31 picks later as we know now. I’m not saying NGS is perfect, but it can guide us a bit to be more informed.

So we can use sliced NGS data to not only rank the classes, but describe them too a bit traits and scores wise. And hey, that’s what I did.

Some findings:

Contrary perhaps to overall belief, the 2026 looks like a very good NGS class overall, but not in the usual “tons of blue-chip freaks” way. Instead, it looks more like a deep, athletic*, high-floor class with though weaker top end players and below-average sized prospects overall.

*More on the athleticism point in a bit

This is a fairly meaty draft class from outside the top 10 to the bottom. There are only 3 players in this draft class with a 90+ overall NGS score, the defined cutoff for a blue chip prospect. That’s the fewest since 2022 (Aidan Hutchinson, Trevor Penning, Breece Hall) and only one more than 2009 (Aaron Curry, Brian Orakpo).

2013 is generally thought to be the worst class in modern-ish history, and in hindsight it may be, but the NGS overall scoring would have given you five blue chip players (Kyle Long, Lane Johnson, Eric Fisher, Luke Joeckel, and Jonathan Cooper).

What about that 2014 class, which NGS scores would have told you was a pond stocked with high end fishes (bluefin tuna?).

That class featured Aaron Donald, owner of the highest NGS score in the database, who yes, the NFL page had him as an “average backup or special-teamer” according to analyst Nolan Nawrocki:

Quick sidepoint: you’ll notice the prospect grade (5.90 for Donald above) you may see on NFL pages isn’t the same as the NGS score data. This is because this grade is based functionally solely on film, and is the scoring of an individual analyst (traditionally Lance Zierlein in the most recent years). They should not be confused and it doesn’t appear they have any linkage to each other.

Looking at everyone else besides one of the greatest players in NFL history, certainly some misses here. Justin Gilbert as the 2nd best player was incorrect, but the Browns didn’t think either it was a massive reach an Gilbert was seen as a top a top 10 player (9th on the consensus big board per Arif Hasan). Shazier was fairly productive, as was Bradley Roby but I think that’s an NGS miss (31st was about right to take him and was 33rd on the consensus board). I’d say Jake Matthews was about right, but wouldn’t call it an NGS hit because the big board was high on him too at #3 overall - if anything it just confirmed his ranking.

You can go down the list yourself here, but the obvious one was Davante Adams, who NGS had as WR3 (should have been higher of course) but was 42nd on the consensus big board and ultimately went 53rd overall. The ding on Adams was that he was not particularly fast or big (which Mockdraftable below echoes) and that he went to a lesser school.

It’s interesting (and forgive me here I’m not a draft historian), Sammy Watkins didn’t seem to have many of the same size concerns as Adams, despite having nearly identical size scores.

Now it’s true, Watkins was seen as more athletic and had objectively better speed and higher RAS and athleticism scores. But what NGS scores can help us turn large levels of subjectivity into some level of objectivity when it comes to college performance. Watkins no doubt had a great career at Clemson and that was weighed strongly in his draft eval (and he has a 91.14 production score). But it doesn’t seem like the production evaluation for Adams was correct. Yes, Adams went to Mountain West school Fresno State but NGS clearly said these were clinically insane numbers to be putting up against any DI competition. There are only six receivers to get a 99 NGS production score:

• Ja’Marr Chase

• Travis Hunter

• Brandin Cooks

• DeVonta Smith

• Davante Adams

• Michael Crabtree

That’s four top 10 picks, Brandin Cooks who went 20th overall, and then Davante Adams. And Cooks only went 20th overall because he was legitimately a very small receiver (DeVonta Smith was small but not to the level of Cooks). My new general rule is if a wide receiver has a 99 NGS production score, take him before the near-end of the 2nd…

Okay so back to the 2026 class, I said it was athletic, according to NGS scoring. Compared with the average of prior class means (these would be better as z-scores but we’ll just use raw numbers sorry):

• Athleticism: +3.45

• NGS overall: +1.94

• Production: +0.17

• Size: -1.56

One thing I think about fairly often with putting draft classes for any sport into context is depth vs difference makers. Difference makers describes the quality at the top, whereas depth is how good the quality is of the class after those difference makers. This may not be a difference maker draft, but it is very solid overall.

2026 has:

• 3rd-best class mean NGS (so the average NGS is higher)

• 2nd-best class median NGS (mean vs median)

• best 25th percentile NGS (so the bottom quartile is better)

• but worst 90th percentile NGS (the top decile is worse of any class)

In all, there is real talent in the middle down to the bottom, but at the top it’s as weak a class as there has been.

It also has:

• fewest players with an NGS score of 80+ for any class (only 25 players)

• only nine players at 85+ NGS, which is 17th of 19 classes

• only three players at 90+ NGS, which too is 17th of 19

This year’s class is not loaded with any elite NGS outliers (though Jeremiyah Love, Sonny Styles, and Kenyon Sadiq are very good), but it is unusually sturdy across the middle to the bottom.

Speaking somewhat further on depth too, 2026 features the lowest standard deviation in production score of any class so far. This isn’t an overly productive class, but it’s tightly compressed without any top or bottom standouts.

The class is deep and balanced, but is not especially explosive at the top.

Subscribe now

Positional strengths and weaknesses

Safety

This class features:

• 1st in average athleticism

• 1st in average overall NGS score

• 2nd in production score

A.J. Haulcy, Caleb Downs, Emmanuel McNeil-Warren, Dillon Thieneman, and the rest of the safety crew don’t quite sweep all the categories, but it’s clearly one of the better safety groups we’ve seen in quite some time.

Interior Offensive Line

There is a bit of catch with this one. Rather than trying to re-tag every historical player who at the combine worked out with guards but ended up as a center, or worked out with centers but ended up as a guard, or any of the two and ended up as a tackle, etc. I did some general re-labeling using prior years roster data to try to find out who worked out with either of the two IOL groups and then ended up as a tackle (I don’t believe this yielded a lot of results but there were some), and then just labeled all the remaining players as IOL. So a fairer system would explicitly label all prior prospects to their correct center or guard tag, then make a projection for the 2026 on if they will be a guard or center (or tackle?) in the NFL. I went a little more simple here. And at the same time, players can start at one position in the NFL then move to another, so it’s even trickier.

In any case, this is one of the strongest IOL groups

• 2nd in athleticism (Jager Burton, Trey Zuhn III, Keylan Rutledge, and Jalen Farmer all in the top 15 of the database)

• 2nd in NGS (22 out of 32 IOL have an overall score of 70+)

• 4th in size

• 6th in production

Cornerbacks

2026 CB ranks:

• 2nd in NGS

• 7th in athleticism

• 6th in production

• 9th in size

By size and athleticism scores, this isn’t a stellar class but they are well rounded when you fold in production to create a solid class.

D’Angelo Ponds is one of the smallest corners by size score we’ve seen and actually shares some similarities with Tyrann Mathieu (who actually worked out with the corners at the combine rather than safeties).

Wide Receiver

• 2nd in NGS

• 3rd in athleticism

• 10th in production

• 17th in size

Really a weird class. Not overly big and in fact, one of the smaller groups we’ve seen, and not overly productive. So why such a high NGS? My suspicion is this is due to the weights for wide receivers and overall score via what drives NFL success (per the NGS team’s definitions of it). Size just isn’t a major driver for NFL success, as receivers are one of those positions that you can find success in all shapes and sizes (literally). We’ve seen DeVonta Smith (68 size score) and A.J. Brown (83 size score) both have nearly identical NGS overall scores (~88).

Offensive tackle:

• 2nd in athleticism

• 4th in NGS

• 12th in production

• 12th in size

Athletic, good overall, but not overwhelming in production. Looks more like traits over resume.

Tight ends

• 7th in production

• 2nd in athleticism

• 19th in Size

• 4th in NGS

Something we’ll cover later, athleticism is a big driver in NFL tight end career outcomes, and this is a very athletic class.

Defensive ends

• 5th in production

• 6th in athleticism

• 19th in size

• 9th in NGS

Rueben Bain Jr not helping out the class here in size, as he’s got one of the 30 lowest size scores for anyone labeled as a defensive end. But it’s not just him, R Mason Thomas and Cashius Howell are both top 45 prospects and both have bottom 30 size scores (same for Nadame Tucker, Trey Moore, and Romello Height too in the size department). A lot was made of Bain Jr’s arm length, but Howell actually has slightly shorter arms than Bain.

Defensive tackle

• 19th of 19 in production

• 18th in NGS

• 10th in size and athleticism

Okay now this also matches the general vibes of the positional class. On the consensus draft board, there isn’t a defensive tackle ranked in the top 20. Peter Woods 77.8 overall score is the 83rd highest DT score and represents the highest in this class.

Defensive line

Because players often move around the defensive line and we’ve got tweeners of defensive tackles that end up as defensive ends and…whatever the opposite of a tweener is in defensive ends that end up as defensive tackles, let’s just create a “defensive line” group by combing DE + DT

• 14th in production

• 7th in athleticism

• 19th in size

• 13th in overall

The best DL classes historically tend to look like:

• Top-5 in production

• Top-5 in size

• At least top-8 in athleticism

But 2026 is doesn’t hit any of those benchmarks (undersized + middling production + decent athleticism).

Pass rusher

Let’s zoom sideways a bit create a pass rusher category, combining defensive end and linebacker (yes not all linebacker prospects are pass rushers but many eventual edge rushers get labeled as linebackers and it’s a skill that is rewarded for linebackers)

• 16th in production

• 3rd in athleticism

• 18th in size

• 8th in overall

Removing this awful defensive tackle group (as you’ll see) and focusing mainly on the guys who are lining up outside the guards shoulders creates and athletic group who is solid but small (edge rushers and linebackers) and lacking production (linebackers).

Running back

19th in production

14th in NGS

12th in athleticism

Another vibe-matcher, this class features a truly blue chip runningback prospect in Love, then falls off a cliff going from Love’s 95.9 to Nicholas Singleton’s 78. For some perspective, Love’s 95.9 score is the highest RB NGS score in the database, Singleton’s 78 is 105th highest.

Linebacker

• 3rd in NGS

• 3rd in athleticism

• 18th in production

• 17th in size

Fairly mixed metrics. It looks a bit traits-driven, though weaker historically in production and size. In a future article, we’ll dive into which individual NGS scores drive overall player outcomes via WAR.

Kickers, punters, fullbacks, and long snappers

Yeah right

Share

Freaks (complimentary)

Alright, let’s look at some outliers

Jeremiyah Love (RB)

The biggest outlier in the class. Love as mentioned earlier has a 95.9 NGS, the #1 RB NGS score in the entire 2008–2026 dataset. His 95.94 production is 9th-best among all RBs. He’s a rare “best in his position" prospect.

Who are the other “best in his position” prospects?

A pretty good list to be on! I’m not sure what illicit substance the NGS model was sucking down when it came to Justin Gilbert but it clearly just loved his athleticism and production. This is a blind spot for the model, as it can’t quantify (nor would you want it to really) professional reliability. Gilbert in college was ultra productive and an athletic player, but in the NFL there were questions of effort and and a lack of consistency.

For what it’s worth, the other highest scoring corners ended up being pretty dang good:

We might bite our thumb a bit at Rome Odunze but I think we can all see him having a decently successful career and not being a black mark on the NGS model 10 years from now .

Sonny Styles (LB)

Coming in with a 94.94 overall score, that’s the 7th-best LB NGS since 2008. Given how weak the linebacker class is in production and size on average, Styles stands out even more.

Kenyon Sadiq (TE)

Sadiq has a 93.32 NGS overall score, which makes him the 3rd-best TE NGS era. His 98.57 athleticism is the 2nd highest (after Zack Kuntz!) in the scoring database.

Jeff Caldwell (WR)

A fun one here, Caldwell’s 99 athleticism score ties him for the highest ever for a wide receiver alongside Parris Campbell (good), Christian Watson (good when not in a cast), Breshad Perriman (mostly not good), and Miles Boykin (not good period).

Those guys though all clocked in with higher overall scores as they were generally more productive in college. But he’ll be one to watch, as all those other players went in the top 100, while Caldwell is currently 182nd on the consensus big board.

Taylen Green (QB)

The signal caller out of Arkansas was given a 99 athleticism score, putting him alongside Kyler Murray, Jalen Milroe, and Tyrod Taylor. Nobody is mistaking Green for a high end prospect like Murray was, but Milroe went 92nd overall and Taylor was picked at 180th. I know it’s lazy but the Taylor comp is just too obvious to ignore.

David Bailey (DE)

Bailey’s 98.14 production score is the 6th highest for a DE prospect and the clear highest of the class (3 points higher than the next player in Love). The consensus board right now is Arvell Reese at #2, Rueben Bain at #6, and Bailey at #7. And IDK man…something is itching in me to prefer Bailey at the end of the day over the other two.

Athleticism Selection Bias

Earlier I mentioned a caveat on the recent athleticism scores and I think it’s become fairly obvious even without having to run the numbers. Players are being far, far more selective on which drills they attempt at the combine or even their home friendly pro days.

Average NGS athletic score

• 2017: ~69.3

• 2019: ~71.5

• 2023: ~72.1

• 2024–2025 peak: ~73.2

• 2026: ~72.9

Let’s just go with the headliner of all the testing drills and 40-yard dash times and participation:

• 2008–2012: ~70–75% of players ran

• 2023–2025: ~45–47%

• 2026: ~58%

We got a minor uptick this year in participation rate vs the prior three years, but it’s a clear dip from when nearly 3/4ths of all combine participants.

Bench press participation rate

• 2008–2012: ~63–70%

• 2025–2026: ~18–21%

Average drills per player at the combine

• 2008–2012: ~4.6 – 5.1 drills per player

• 2022: ~3.3

• 2023: ~2.5

• 2024: ~2.47

• 2025: ~2.34

• 2026: ~2.83

Again we saw a slight uptick in overall participation this year from a total drills perspective too, but still down drastically vs history.

And this data makes sense from a narrative point too. Prospects are being selective for which drills they are good or bad at. They are testing in the ones they know they will perform well at, and skipping the ones they know they won’t. Which I wonder if the sub-optimal way for a prospect to approach this?

Like, if you are seen as a slow runner, does not running in the 40 help? If your 40 time is slow, okay, that confirms the already expected figure…does that actually drop your stock? And there is upside too, in that if you come in faster than expected, your stock shoots up. Sure the inverse can be true, a fast runner can only confirm he is fast or hurt his stock by running slower. I suppose a prospect has to weigh the impact of a counter-narrative on the upside vs downside.

I think if I were a perceived slow runner, I might run the 40 and shoot for the upside. If I am perceived as a fast runner and particularly that is a lot of my profile, I might skip the 40.

There is also no correlation between number of drills performed and athleticism score, which shows the selection bias further. Bad results never enter the dataset.

There’s also a bit of selection bias in the NGS athleticism score input methodology, as NGS allows for the better of the two scores of combine or pro day (though I think it’s rare a player does the same drill at both?).

This does taint the athleticism scores a bit for the recent classes. Focusing on 2026 again it has:

• 3rd-highest athleticism class

• few elite NGS players

• tight production distribution

• worst size class

This screams a heavily filtered class that has fewer bad athletes showing up in data, fewer extreme outliers (because fewer full testers), and a tighter middle distribution.

Let’s stick on the athleticism point just a little longer and split the group out between full testers (4+ drills performed) and selective testers (2 or fewer drills):

Full testers

Average number of drills ~4.8

Average athleticism score ~70.1

Selective testers

Average number of drills ~2.0

Average athleticism score ~73.4

And this gap is a modern class phenomenon on athleticism score gaps:

2008–2012:

Full testers - Selective testers has a gap of ~0–0.5 points

2016–2019:

~+1 point

2022–2026:

~+3 to +4 points

And we can look at drill level biases too

• Players who skip agility drills have higher average athleticism

• Players who skip bench have modestly higher average athleticism

Players who only do speed/jumps have the highest scores

Average 40 yard dash times of the bottom quartile:

2008–2012: ~4.78

2016–2019: ~4.73

2023–2026: ~4.68

Average 40 yard dash times of the top quartile:

2008–2012 4.43

2013–2017 4.42

2018–2020 4.41

2021–2022 4.40

2023–2026 4.39

Standard deviation of 40 yard dash times

2008–2012 0.182

2013–2017 0.171

2018–2020 0.163

2021–2022 0.154

2023–2026 0.139

So the average 40 yard time of the bottom group has dropped, and the variance has compressed, but not because everyone is getting faster, as the top quartile has remained mostly the same. It’s simply bad runners stopped running.

It’s possible that even though athleticism scores are inflated, it might not be that big a deal. You’d have to optimize each drill for each position and how that impact future performance outcomes. Like, the broad jump for a wide receiver doesn’t really matter, so them not doing that drill probably doesn’t matter. Even if a really poor jumper doing it would bring down their presumed athleticism score relative to others, it wouldn’t really matter on the future success outcome. NGS I believe does this adjustment already when it factors in which drills matter. Yes, it still does have some upside bias, but lack of drill participation on it’s face might be a little overblown.

Ideally what we’d want to do is rebuild everyone’s NGS scores based on their combine/pro day participation data, but 1) that’s an exercise for another day and 2) I don’t know how much of an impact that would really have. Would dropping athleticism scores by 2-3 points really do much?

NGS-implied draft spot

One last exercise before I leave you alone for the day. What if we can take NGS scores and create a sorta implied draft spot based on historical data like a players actual draft pick, NGS score, and where players with similar scores and positions went. We can then use that to try to find guys who went earlier/later than their NGS scores said they should.

Biggest NGS implied pick steals for players taken top 50:

And the inverse, guys who NGS scores taken in the top 50 that said they should have gone later (we are ignoring QBs since they naturally get pushed up the board):

Let’s expand out a little bit and do the same but for the top 100. The “risk” of being wrong against NGS for an overdraft is lessened because it could be less capital used, but on the opposite, the reward is greater.

NGS steals in the top 100:

And NGS reaches in the top 100 (again QBs excluded):

Tying it back to the 2026 draft class, what if we used that same implied NGS pick concept but rather using the pick the player was taken at (which we obviously don’t know yet), we use their ranking on the consensus aggregate big board.

Here is for players in the top 100 of the consensus board, whose NGS score implies they should be higher:

And again the inverse:

The two that really stand out here are Carnell Tate and Francis Mauigoa (honorable mentions for Terrell and McCoy too).

Let’s talk a bit as to why Tate isn’t quite as loved by NGS scoring relative to the analyst community. Most wide receivers taken in the top 10 were highly productive in college, with average productions score of 88.8. Tate is at just 78. The last WR taken in the top 10 with a production score at 78 or below was Darrius Heyward-Bey (taken 7th overall).

Here is every first round receiver with a production score <=78

Yes, Beckham Jr is screaming at us and Egbuka has been good too in his limited career, but they like all the other receivers on the list had higher athleticism scores than Tate (and remember that Tate is playing in an era of higher athleticism scoring too):

Tate has by far the lowest athleticism score for a 1st round receiver who scored not great in production also. The only other comparables to Tate are Mike Williams and Demaryius Thomas, a trio of lower athleticism guys, though Williams was highly productive at Clemson, whereas Thomas was not (at Georgia Tech).

I think this is a clear example of the gap between the film analysis and the stat-based analysis. If Tate is to defy the stat side of things, it’s going to be on things that stats can’t innately capture like nuanced route running, projecting maybe some addition to his size, and his ball tracking abilities.

Future pieces over the next few weeks will keep diving into this data in as many ways as we can.

(self-editors note: this piece was supposed to be published last week but was not, so references to “Sunday” are slightly dated)

The Chiefs 2025 season ended on a normal for December cold Sunday afternoon in Kansas City. Well, actually, visions of it ended on Thanksgiving at AT&T Stadium with a loss to the Cowboys. It ended in practicality last Sunday night at Arrowhead in a loss to the Texans. It ended in reality on Sunday, and instead of waking up from the nightmare, the Chiefs sunk further into the dark sleep state.

You see when a season starts, you have dreams of trophy hoisting. And even if those dreams just end up tricking you and are eventually nightmares, at least when the season ends you wake up and escape. Unfortunately, the 2026 Chiefs may still be tortured in slumber.

About an hour or so after the quad zeros hit on the clock for both the game against the LA Chargers and the 2025 season, reporting dragged the team down deeper into levels of sleep terrors. Patrick Mahomes had torn his ACL. 2025 was officially lost for good. 2026 now had some mystery hovering over it, before the calendar even flipped.

But I’m not here to talk about timetables for injury returns. I’m not a doctor. I didn’t even take biology or chemistry in college, and if I did, I would have failed both. Maybe Mahomes misses no time in 2026. Maybe he follows the Kyler Murray path, hurt in December, out until November the next year. Maybe somewhere in between. It’s worth considering for the outlook on the 2026 team, but it’s impossible to plan for all scenarios because they all have different independent paths. If Mahomes misses no real time and is good to go by training camp-ish, season in business as usual. If it’s a few weeks into the seasons, it’s business somewhat as usual depending on the schedule (they play the AFC East and NFC West next season - two tough divisions) and which opponents line up to those windows. You can imagine the NFL won’t likely give the Chiefs a marquee matchup in week 1. If he’s out for a large chunk of the season, it’s business as unusual (is this a term?).

You could even argue that the Chiefs season was on the path to being over before the season even started. This wasn’t a team that was decimated by injuries, creating an uphill climb.

In 2024, the Chiefs were one of the least injured teams by SIS Total Points, ranking 29th. This year that has risen from 47 points through all of 2024 to 71 points through early-December, putting them at 19th most in the league. A jump for sure, but not a crippling amount, and Total Points considers the impact of the player too, not just the raw count.

Look at the teams at the top here, most of them have losing records (Cardinals, Bengals, Commanders, Giants, and Falcons) or are treading water (Dolphins and Bucs). But there are three resilient teams. The 49ers not only have missed the most Total Points, including their starting QB, star linebacker, and star defensive end, but are 10-4 and in the thick of the fight for the NFC one seed (Kyle Shanahan coach of the year). The Bills aren’t quite in the same fight as San Francisco due to the Broncos having a strong hold on the bye, and it remains to be seen if they even win their division, but they are solidly in the playoffs on the AFC side. And the Chargers, coming off a victory that knocked out our 2025 Chiefs we are speaking of, are likely a playoff team and are a win away from sweeping their division.

One more way we’ll look at it, just by raw dollars missing from the field via injury or discipline (per OTC)

I’m walking a little crooked down this commentary path and not getting to the point of the piece, but I just don’t want the narrative to be that the Chiefs season was derailed by injuries. Injuries didn’t put them in an uphill battle, the healthy team did that. The injuries came when they were already at the bottom of the hill.

So to wander back to the middle of the path here, the question is what caused the 2025 Chiefs to be at the bottom of that hill, looking up at it ten weeks into the season.

In business (or at least in business school), there is a thing called a pre-mortem, where you imagine how a project could fail before you even begin to start working on the project. The patient has already died, what killed them? It’s supposed to help you think not about what could go wrong, but starting from failure and working backwards.

Had the Chiefs done this in let’s say April 2025, they maybe could have seen the issues that would plague them, instead of thinking about what issues could plague them.

But it’s not fair to run a pre-mortem today, when we know the patient is dead and what killed them. There will be plenty of 2025 Chiefs post-mortems written over the coming weeks and months. The honest ones won’t blame injuries (at least not as a core factor).

Let’s talk about the future, at least, the future that is ~9-12 months, if that is any sort of a future. If I told you you were going to die in your future, you probably don’t think/hope it’s in the next 9-12 months or at least you have a few questions for me. But in any case, let’s call it the short-to-mid future, because when you plan for a roster, you aren’t just planning for next year’s roster, but the next years rosters.

The easiest starting point is the cap and roster as it looks today, knowing that it will change.

It’s a long way still until the new league year in March (11th), but the Chiefs are currently $43M over the cap per Over the Cap (2nd lowest to the Cowboys) and have the 3rd most snaps up for free agency, right behind the Bills (also $3M over the cap) and Brown ($7M over).

Some of the pending free agents include names like Travis Kelce (who has played 80% of the Chiefs offensive snaps and just might be retiring), Hollywood Brown (50%), Bryan Cook (76% of defensive snaps), Juju Smith-Schuster (51%, Jaylen Watson (97%), Leo Chenal (52%), Charles Omenihu (55%), Kareem Hunt (49%), Isiah Pacheco (39%), Tyquan Thornton (37%), Jerry Tillery (32%), Derrick Nnadi (32%), and just for a final little bit, even their punter Matt Araiza is technically a free agent (it’s an exclusive rights free agency so he’s not really going anywhere but we’re having a chuckle).

In some blessing, the names above aren’t exactly like those late-2010s Patriots who over a span of a year or two had just about everyone of importance for the second dynasty run due up for free agency, but that’s still a lot of snaps. And a few of those guys have only ever been coached by Andy Reid and Co., so there is some institutional memory being possibly lost too.

Back to the cap, the all the red ink you see over at OTC is real and at the same time, an illusion. The Chiefs are as it stands $43M over the cap (which is a lot) but the cap, as it always has been, is temporal and malleable. With one click of a mouse, the Chiefs can send an email to the league office and clear that right up by restructuring Mahomes contract, which itself saves $44M in cap. This is exactly what his contract was setup to do (all QB deals are this way now).

But let’s pause for a second before we play games over at OTC and separate could and should. The Chiefs could push Mahome’s 2026 cap hit forward, and it takes his 2028 figure up to $85M, but another restructure when that bill comes due drops the cap hit down to a reasonable $44M. With the structure of his deal, 2027 looks like the final years where the cap hit on paper starts with a 7 or 8 after restructures and we’re looking more at 5s and 6s. The Chiefs should do this, the injury makes no impact on this decision. The contract structure exists for this specific purpose. To not do it is to deny the reason you set things up specifically for.

Now on to the shoulds. Chris Jones is cap figure is staring us all right in the face with infinite and fearless eyes. Restructuring his deal saves $22.5M, but takes his 2027 cap hit from $38M to $49M. Now, that $49M only depends on if they keep him for that season, but to cut him would cost them ~$41M in dead money for 2027 if not (the money the pushed down the road + the immediate cap hit of the 2027 and 2028 bonus prorations). Is that the best recourse? To create 2026 cap space via a 32 year old defensive tackle, who while still has the capacity to destroy an offensive lineman, seemingly only wants to do so for for maybe 5-10 snaps a game and is on pace to have his lowest total pressure, pressure rate, and pass rush win rate of his career. Perhaps the Chiefs would be best to just like his 2026 cap ride (or only restructure a bit as needed) rather than the full course.

But the two “obvious” levers to pull are Mahomes and Jones, which could free up $66M in cap, taking them from deep in the red to fairly into the black at ~$23M, which if no other team made a move, would put them around the median. But is isn’t that simple.

The Chiefs are $43.7M in the red, but that doesn’t consider that they have a near league low 37 players under contract for 2026. What we care a bit about is what is called effective cap space, as the remaining 14 spots have to go to someone and by FSLA law those players have to be paid. Haven’t heard of many unpaid internships for NFL players, but the Bengals are considering it. Also included in the effective cap space idea, the Chiefs have to sign their draft class, and in true Loser’s Curse fashion, the better your pick, the more those players are going to cost you. You see, one of the benefits of picking 32nd instead of 1st if you get $7.5M in cap savings on your first round pick, amongst other things.

The other last elephant in the lever pulling room is Jawaan Taylor, who if cut, would clear a cool $20M in cap space, only leaving a reasonable $7M behind. And let me ask: has there been any talk or whispers about Taylor being a cut candidate? The fans don’t like the contract, sure, but when he’s not committing penalties he’s been a serviceable pass blocking offensive tackle. And given all the injury attrition the Chiefs saw on their offensive line this year, it’s reasonable to think the Chiefs front office isn’t going to cut their healthiest offensive tackle (and lineman outside of Creed Humphrey). The other options seemingly are after cutting Taylor to 1) give the job to Jaylon Moore and his $18.7M cap hit 2) let UDFA Esa Pole, a player the Chiefs cut not long ago, get those snaps 3) draft a tackle in the first two rounds…again or 4) sign a cheap vet in free agency to help with any of the prior three (which doesn’t help on the cap side of things necessarily). Moore would actually bring about $8M in cap savings, so it’s possible, but the Chiefs probably want depth here overall between Taylor and Moore. I don’t think it would be wild if the Chiefs held pat (no pun intended) on both here.

But we also said that roster planning is not a short-term thing, you have to think about years forward too. The Chiefs currently don’t have any offensive tackle under contract in 2027 outside of Josh Simmons (and Kingsley Suamataia who we are all considering a guard going forward). Is there a development tackle somewhere after day one in the draft? It’s worth considering, but at the same time, that’s something that either really bad (plenty of snaps to give to him) or really good (can wait awhile to develop him) teams can usually afford. I’m not sure the 2026 or 2027 Chiefs are either right now.

The rest of the players with true cap savings:

Drue Tranquill: $6M - I don’t know hard for me to see him being cut for such small savings. He’s been good and is on the final year of his deal anyways.

Mike Danna: $9M - yeah…thank you for your service Mike. But to be fair, the Chiefs defensive line depth is really bad, especially if you cut Jones. Maybe he’s more of a re-worked deal than a see ya later.

Kristian Fulton: $5M - seems like the Chiefs front office is going to give up the ghost on this one, other than Mahomes, this is the most obvious move.

Noah Gray: $4M - I mean sure but why? That’s couch cushion money and Gray probably ends up a Chiefs lifer (okay maybe he spends a year or two elsewhere Demetrius Harris did).

That’s really the list. While the below is just looking at simple restructures (converting unguaranteed money into guaranteed money under existing contract length, no void years), the Chiefs would still have the lowest cap space in the league if every team pulled their simple restructure levers.

To really have some cap in their wallet (not cash here, which is separate), you are going to be looking at a Mahomes restructure, something with Chris Jones’ contract, saying goodbye to Taylor, Fulton, and Danna. Pulling those levers (let’s only restructure half of Jones’ deal) gets us to ~$45M in cap and ~$26M in effective cap (I’m going to just ignore the 3 roster spots that would have to be filled with those cut players at some amount of cheaper money). That would be good for 12th in effective cap space (again assuming nobody else makes any moves).

Chat, who are we bringing back with that bit of money? I’d vote for Bryan Cook, Jaylen Watson, Leo Chenal, and Tyquan Thornton. That still leaves a lot of snaps to fill but this isn’t the US government, we don’t have monetary sovereignty and can print our own money.

Let’s pop over to Tankathon and look at the Chiefs 2026 picks, which will help them restore the ros-

That’s…not a lot! Despite currently slotted to pick 12th, the Chiefs have only the 20th most draft capital per the Fitzgerald-Spielberger Trade Value Chart, between the 49rs (picking 27th) and the Cowboys (who have two first round picks but don’t pick as of now between pick #24 and #113). The Jaguars are likely going to win their division, have no first round pick, and still have more F-S Value.

Before we hop to draft needs and thoughts, one thing we should touch on is a way to create draft picks without losing games, and that is via trade. There isn’t a ton on the Chiefs 2026 roster that is 1) tradeable or 2) going to move even toy sized needles you had in your fake doctor kit as a kid. They aren’t trading Creed, Trey Smith, or Rashee Rice. The only “real” candidate Trent McDuffie, which would be a re-run of the L’Jarius Sneed trade from 2024 (though they are different players). Sneed netted the Chiefs a third rounder a year later in the 2025 draft (ignoring the 7th round swap). And while he isn’t an elite outside corner like we saw with Sauce Gardner getting two firsts, McDuffie is worth more than Sneed and should return a 2nd round pick and change. How about this as a comp: February 23rd, 2018, the Chiefs trade Marcus to the Rams for a 2nd and 4th round pick (again let’s ignore the 6th the Rams got back just for a second). That 2nd rounder the Chiefs got back was the 63rd overall pick, which is currently held by *checks notes* the Rams! Let’s repeat history and do the same swap. If we give the Chiefs the Rams 2nd and 4th round picks (the Bears have the Rams 4th but this is all make believe anyways), they jump all the way up to 6th overall in F-S Value, right between the Steelers (who have a million 3rd and 4th round picks) and Giants (who have quality over quantity, as they currently own the 1.1).

Reminder: we are trying to fight between coulds and shoulds. The Chiefs *could* trade McDuffie, but should they? This would be the cornerback room without him, before any re-signings: Nohl Williams, Christian Roland-Wallace, and Kevin Knowles II. That’s it. That’s the list. Those are your corners heading into free agency. A small group of guys with a large amount of little experience or production. A lot of the benefit you get from trading McDuffie (draft capital, cap space, future cap space), you are just going to have to use some to fill that newly created hole from the trade, and you just can’t expect to plug it with a replacement for a great corner like McDuffie.

But let’s live in that world and trade McDuffie and run a mock draft (note: I had to “trade” Hollywood Brown to the Bears to make the 4th round pick work, so ignore that). I might do something like this:

Focus more on the positions than the players, but the Chiefs need to be spending a lot of resources in the upcoming draft on defense. I hesitated even taking a tight end but they need someone else there alongside Gray, as I assume Kelce will retire.

Mock drafts though are just a single pass, just one run down the old Monte Carlo simulation where there are millions of other path dependent outcomes. Let’s run it one more time

I want to take a side step here for a second and address gorilla in the room (who keeps putting these wild animals in rooms with us?)…in both trials here, Jeremiyah Love was available at the 11th pick and we did not select him. It’s not because I couldn’t find the button or his ancestors killed my ancestors because the king ordered it. The Chiefs have NO business taking a running back in the first round. Allow me to just summarize with some of my own tweets (it’s really eye rolling to do this I know but it’s already in a nice and neat format)

There isn’t much evidence that NFL teams can separate the good RBs from the bad ones, including separating the ones who are good at pass catching or not. Love seems like a great prospect, but one the Chiefs should have no business spending ~28% of their draft capital on (and without the McDuffie trade it would be 35%). And this isn’t even about the player or the position itself but the head coach.

Andy Reid just doesn’t want to run the ball. Like, ever. Especially with Mahomes at QB

And this is not just a Mahomes thing, even with Alex Smith at QB and Kareem hunt lined up in the backfield, it only changed things slightly.

With Jamaal Charles instead of Hunt, it was more of the same.

Going all the way back to his time in Philly, when he had LeSean Mccoy and Brian Westbrook to hand it off to, it was still early passing.

Andy Reid does not want to run the ball. He’s never shown that he wants to be run-first, let alone far off from an even mix. And that goes double since he got to call plays for Mahomes.

And there for me is two crucial points. First, do you want Mahomes throwing the ball less? Especially on early downs? Okay, maybe for the first bit of his return from the torn ACL but over the next 4-5 years? Whose hands do you want the ball in when it’s time to put up points in neutral scenarios and early downs? Second, if Reid isn’t going to give the rock to someone like Love, why spend the significant resources there? We clearly know that even if Reid has a great back to hand it to, like hall of fame talent type guys in McCoy, Westbrook, and Charles, it didn’t matter. Even with a better worse quarterback (sorry Alex) and a better running back, it only dragged the Chiefs offense more towards the middle. We know Mahomes is better than Smith by some bit, but are we sure Love is as good as Westbrook/McCoy/Charles?

It’s tough to argue that Love adds a ton to the passing game, as he’s good not great receiving back. If you want a pass catcher, take a receiver. If you want a rusher, take a running back later in the draft. If you want both, you are betting a lot on your ability, which is against the trend, to find the good RB and take him early.

And this is all beside the point that the Chiefs just do not have the capital, be it cap or draft, to spend a significant singular resource on a RB. This isn’t a “luxury” pick like CEH in 2020 when the Chiefs had a fairly full roster, an MVP QB on a rookie deal for another year (plus a cheap year after), coming off a Super Bowl win, with a lot of the core players still younger and under deals.

Just none of it lines up to a smart decision, no matter how good Love is or isn’t, and we focus on process over results.

So what position should interest the Chiefs in the draft? Let’s avoid the word “target” because the Chiefs roster is empty enough that they can take the best player available at a few positions and be happy with it. This isn’t a roster and 2026 isn’t a year where they just need to shore up one or two positions for an already deep roster.

Quarterback - I’ve long opined that teams should every 3-4 years select a QB late in the draft, just to save themselves a few million in cap that they pay a vet QB. Some vets are making $6-7M per year. I’m a believer in the old Tom Moore quote of “Fellas, if 18 [Manning] goes down, we’re f—-, And we don’t practice f—-.” Without Mahomes, the offense is in trouble that even a vet can’t save (Chad Henne miracle excluded). I’d consider one in the 7th but considering the Chiefs don’t have a 7th round pick, it’s going to have to be a UDFA here.

Halfback - To clarify my earlier texts, the Chiefs do need to add to their RB room, 100%. But they need to use several lower value resources on it, not the 11th pick.

Chiefs should use a day 2 or 3 pick on a running back, while also bringing in some guys on cheaper deals, avoiding spending a big bag on someone like Breece Hall or Rico Dowdle or Javonte Williams.

Tight End - This has been a popular pick, to find a Travis Kelce replacement, but as we all know, that’s far easier to say and dream of than to do. What makes Kelce special isn’t that he was some freak athlete or a thunderous soul crushing truck sticker. He won less on the physical side of things and more on the mental aspect. Elite in finding the right space on the field to occupy, connecting with his quarterbacks, and leveraging angles to know which way to turn when a defender tries to tackle him after catching the ball. That just simply isn’t a replaceable skill. This is why I think it’s worth warning that signing someone like Kyle Pitts seems enticing, but he’s not a Kelce replacement. They operate completely different ways (and Kelce also is a better blocker). Kenyon Sadiq from Oregon pops up on Chiefs fans radars, but he’s not a Kelce replacement either. Nobody is, and Chiefs fans need to realize Kelce is impossible to replace exactly and to also not put expectations that any tight end taken has to live up to that standard. On the merits of drafting a tight end, it’s not quite a steep a fall off as receiver, but there is a small step in separation between first rounders and second.

If the Chiefs want to add to their pass catching game, they should probably just take a receiver in the first, then see what tight end is still hanging around in the next few rounds. Though offense isn’t the side of the ball I’d focus on overall, if a pass catcher is too good of value to pass up, then fine with that. But also keep in mind, the league is trending more towards heavier personnel and more balanced run/pass mixes, and the importance of a tight end that can also run block hasn’t been higher in some years.

Wide Receiver - I really go back and forth on this one. The Chiefs have spent a good bit of capital over the years at the wide receiver position, multiple day one and two picks. And while the room is better in 2025 than it was in 2023, it maybe still hasn’t scratched the itch and left everyone feeling great about it. Rashee Rice is a good high end #2 receiver, and Xavier Worthy has his merits as a complementary/gadget guy…but this is all just more of what the Chiefs have had over the years since Hill was traded away, and it’s become less effective with Kelce getting older. You could see on Sunday the Chargers just didn’t really respect any deep threats from the Chiefs offense, especially when Worthy went off the field and they just started playing only man coverage.

If the Chiefs want to add to the receiver room (which as of now is set to lose Brown, Thornton, and JuJu Smith-Schuster - a combined 135 targets through week 15), there are options all across the first few rounds, spread pretty evenly.

(via James Foster’s 2026 Draft Site)

One thing though that does keep ringing in my head, the Chiefs need to mix it up if they do take another wide receiver early in the draft. No more small guys, no more super speed guys, and no more YAC-centric guys. They need to pursue a body type and play style that compliments the speed and agility they have on the roster. Defenses just don’t respect the Chiefs receiving group on anything beyond 8+ yards down the field. In part because of the receivers, but because Mahomes has struggled on deep throws post-Hill.

What this might mean is that they need to avoid guys like Jordyn Tyson, Makai Lemon, and KC Concepcion. Instead, someone like Carnell Tate is a better fit for the roster. This isn’t a knock on the first three players, they could very well be good. But the Chiefs need to find something different to pair with Rice (can’t win down field, good in the intermediate and YAC) and Worthy (can’t win down field, good over gadget and short schemed plays). Or even wait for someone like Georgia State’s Ted Hurst sometime in the 2nd or 3rd.

Defensive Line - Another position that the Chiefs have sunk some draft and cap capital into without an equal return. Four 1st/2nd round picks have been used on the defensive line over the past four drafts, and it seems like in most weeks if Chris Jones isn’t winning, then there is no organic pass rush.

This upcoming draft is filled well with great prospects on the DL when looking at the consensus board. That might make you say “well then you can wait on taking a defensive lineman” which brings me to one of my favorite draft topics: depth vs difference-makers.

Depth is in the middle, not at the top. Depth means there are a good group of players who are starting caliber players, but they aren’t difference makers. No one is going to say “well we could get a top 10 players in the class at edge rusher, but we could also wait and get the 40th best player instead.” Gaps in talent exist, and those difference-makers matter. If anything, a deep class calls more for doubling up, as you are leaning into the strength of the class.

The end goal in drafting players is to acquire above market value players at below market value prices. Premium positions have wider gaps in those valuation arbitrage. And further more, they don’t often become available on the open market in free agency, and when they do, you are paying for them either in a trade, cap, or both in trading for a great player and extending them.

One more point on depth, if you choose to take a player later because of depth, you are by definition passing on the idea of best player available. Going BPA, which most folks will tell you in the best way to operate in a draft, means taking that potential difference-maker, not passing on one if the value aligns. You draft for value, you sign in free agency for need.

Linebacker - We can argue all day on if the Chiefs should have made Bolton one of the five highest paid players at his position. Even though the anti-Bolton crowd (which it pains me to be in as a Mizzou fan) has our issues, it’s clear the Chiefs and Spags trust and value him as the leader of the defense. It’s hard for them to let a player like that get away from them like a running back in space against Bolton. He’s going to be around for at least one more year in 2026 given the contract. They are set to lose Leo Chenal, who has turned into a core swiss army sorta guy lining up all over the box and contributing on special teams. I think he’s cheap enough to bring back but he doesn’t seem super high on the totem pole when it comes to where to spend their limited cap dollars. The Chiefs still have Bolton and Tranquill for 2026 (Tranquill is a FA after that season). Chiefs recent 5th round pick Jeffrey Bassa had some pre-season hype, but he’s rarely played on defense and is not a replacement for Chenal’s style there. I’m operating as if they bring back Chenal, which means they aren’t taking a classic linebacker early (we can argue how to classify guys like Sonny Styles and Arvell Reese).

Defensive back- Certainly a team need here, especially if they are entertaining the idea of trading McDuffie, which would presumably happen prior to the draft to maximize pick value. I actually think this could be a sneaky leader for the Chiefs first pick, if you are into that whole betting thing. Maybe it isn’t sneaky but it feels like a lot of the first round mock drafts have the Chiefs taking an RB, WR, or a defender in the box. The Chiefs safety room is on track to be wiped out as well, losing their starting free safety in Cook, his backup Mike Edwards, and sorta all around DB/ST guys in Deon Bush and Nazeeh Johnson. That leaves only Chamarri Conner (who has played more slot corner this year) and Jaden Hicks (who is more of a strong safety) left on the roster as it stands. I doubt they go safety with their first pick because 1) Chiefs seem to love premium positions in the first round and 2) outside of Caleb Downs there isn’t really a safety looking to go that early.

Kicker - Sorry, typo

So where do the 2026 Chiefs go from here? Well it starts with coaching decisions. And this isn’t to say that the Chiefs should fire Andy Reid…they shouldn’t. But they would be well served to step outside their ecosphere a little bit and bring in some sort of fresh face from these adaptive offensive regimes. Think about poaching someone from San Francisco, the LA Rams, Minnesota, Miami, Green Bay, or Chicago. The NFL rules are tricky for coach hunting, since a team can block any move that isn’t a promotion. So the Chiefs can’t steal Mike LaFleur from the Rams just to make him their passing coordinator. If Matt Nagy does presumably leave (his contract expires in a few weeks), then that is easier to grab someone from the outside since you can hire anyone that isn’t either a current OC or HC to be your new OC. But what the Chiefs need to avoid is just promoting Joe Bleymaier (pass game coordinator) or David Girardi (quarterbacks coach) to OC. Nothing against those guys, just that there needs to be some outside voices brought in.

On the roster side, it’s obviously they will restructure Mahomes’ deal, it just a question of how much and in combination with who else they cut/restructure. They can’t get out of the red with just his alone, but they don’t need to do all of his. You could restructure 70% his deal, cut Taylor, and then restructure Trey Smith and Creed Humphrey (both are going to be sticking around for awhile) and you are at ~$20M in cap space without having to touch Chris Jones’ deal at all. But $20M is a very small number when we are just talking strictly about cap space and not effective cap space. Even with those moves, the Chiefs would be at 38 players on the roster and $16M needed to be set aside to fill it out with just NFL minimum salaries and draft picks, so we are really talking about a number closer to $5M. And that is without signing or re-signing a single free agent. It seems like they will have to make some decision on Chris Jones.

Here is a bit of a starting point via OTC ( and remember the cap figures are not effective cap on this page).

This gets you to $36M in cap space (or ~$15M in effective cap space) without creating any huge holes in the depth chart. It does mean that the Chiefs will have to figure out someone to be the swing tackle (Esa Pole?) since I wouldn’t rely on Simmons and Moore as my 17 game starters and Wanya Morris just isn’t happening. But shedding Danna and Fulton is negligible and a net positive for the cap space vs their contribution.

The real question is is 2026 a rebuild or reload? I lean a little closer to rebuild than reload. Not even considering the impacts of the Mahomes injury, the roster is turning a bit into a stars and scrubs mold, with the stars maybe being more like just very good rather than elite. This isn’t the 2020/2021 Chiefs where you still have Mahomes, Kelce, Hill, Jones, Thuney, all in their primes as clearly top 1s or 2s in their position, next to guys who are top 10s in their positions like Sneed, Fisher, Schwartz, and Mathieu. Plus you had Humprey and Smith on rookie deals, and Bolton wasn’t make Bolton money. Not to mention solid contributors in Thornhill, Ward, however you want to categorize Frank Clark and Willie Gay Jr, etc. I even liked Mecole Hardman for what he was.

No, the roster is set to be a bit more bare now, for reasons we can spend the next seven months arguing over. So to me, it’s more like a rebuild than fixable in one offseason. And I mean it in a sense not that the Chiefs can’t compete for a title in 2026, just that it seems unlikely that 2026 will be a dynasty-like roster. It’s very possible they even win a Super Bowl in 2026, which would be the starting point for any dynasty. But it won’t be because the roster is suddenly flipped and become the best handful in the league. Instead, it’s because they have Mahomes at quarterback, and for most cases, all you need is the QB and you are like 60% of the way there.

For those of you who follow me on Twitter, you’ll know that I like to find historical team comparisons based off of teams with similar DVOA components. The way it works is that using every team season in the historical DVOA database, I look at four components of DVOA: total (offense + defense + special teams), offensive DVOA, defensive DVOA, and special teams DVOA. Yes there is a little overlap and double counting there, but adding the total DVOA helps anchor in my opinion the team’s talent rather than trying to find my own weightings of the three non-total DVOA components. What I do is I turn each component into a z-score where the “mean” is the team I want to compare the rest of the other 1,200+ individual team seasons to, and the standard deviation is the standard deviation of that component historically. What this does is it matches every team in the DVOA database component-by-component to the component of the team we are looking at (the 2025 team we are looking for historical comparisons to) and shows how far away that team’s component is to the comparison team.

The benefit of using z-scores is that since we include standard deviation, we can create a normal distribution of each component around the comparison team and thus “standardizing” the component. So a historical team with a 1.5 z-score on offensive DVOA to our comparison team means that it is 1.5 standard deviations away. It’s a nice way to make everything sorta even and adjusting for the volatility of each component rather than just subtracting the raw component from the historical component or something.

Then after turning all four components into a z-score, we turn them into absolute values, and then add them all up. We use absolute values because we want to be a little flexible on not nitpicking between a team with a -0.2 z-score and a +0.2 z-score, which both in our purposes are the same. Both would just have a +0.2 z-score. Ideally we’d use like RMSE to punish for distance but I’m just simply not that good at math or Excel to do that.

Once we’ve summed all the four component z-scores up, we’ve now got a singular z-score. I’ve chosen to just equal weight all four components. Ideally we’d do some sort of weighting, especially down-weighting special teams, but I haven’t figured out or tested the right balance here. DVOA itself isn’t weighted by component, it’s just the offensive DVOA - defensive DVOA (since better defensive DVOA is negative) + special teams DVOA. A good bit ago, Football Perspective tried to estimate team wins by weighting DVOA components, using what would sum to 50% offense, 36% defense, and 14% special teams (though they broke each component down by pass and run and that’s too much for me to do manually for each team every time). I’ll try that weighting here in a moment.

For example though, with our equal weights, here are the historical teams by z-score to the current #1 DVOA team Seattle Seahawks.

The lower the z-score the “better”, in that the sum of all the z-scores add up to smaller amounts and thus, in theory, are closer sum of component teams to the comparison team in Seattle. This is a pretty good top five in the 1996 Packers (won SB), 2013 Seahawks (won SB), 2000 Titans, 1995 Chiefs, and 1991 Commanders (won SB). The list also include a few more teams that either made or won the Super Bowl when you get a little further from ~2 z-score.

What is a good z-score total? I’d say ideally <1 and there are at times some z-score that even get down to <0.5, but typically I like ~2. For instance above has the 2000 Ravens, who won the Super Bowl.

Historically from a percentile and rank perspective, both teams are one of the ~120 best teams by total DVOA and both have had top 25 defenses by the same criteria. However, offensively the Seahawks have been much, much better, marking as one of the 225 best offenses so far. And to add a point of clarification, the historical team is for the full season, the comparison team is through today. So we are comparing a team that played 16 games with a team that has only played 13 games. Not a huge difference but it’s worth pointing out. Unless something goes drastically wrong for the Seahawks over their next four games, it’s not going to change their overall DVOA by much (and of course the 2000 Ravens DVOA won’t ever change unless there is an update to DVOA).

But let’s look at the top two team historical team comparisons to the current day Seahawks.

All three teams above here are top 10 in total DVOA all time historically, and all three 99th percentiles. All three had great offenses, very very great defense, and great special teams. The 1996 Packers were coincidentally coached by future Seahawks coach Mike Holmgren. Not to compare Sam Darnold to Brett Favre, but the year prior, Favre had his first real breakout season where it all came together at age 25, then the next year at age 26 won MVP (his first of three straight) on the way to the league title. Sam Darnold had his first breakout season where it all came together last year (at age 27 though), and are hoping this year to be a Super Bowl winning QB like Favre.

I might at first glance say the 2013 Seahawks from a component perspective really are the closer comparison. They are a closer fit overall to the 2025 Seahawks, as the 1996 Packers offense was a bit better. Special teams here getting an equal weight is what s driving the z-score distance a bit.

Using the previously referenced Football Perspective weightings (50% offense, 36% defense, 14% special teams), we get this instead:

Much lower z-scores but also we’ve completely removed the total DVOA component…which might not have hurt? The 1996 Packers still remain the closest, and their z-score actually gets cut by more than half. The 2020 Saints and 1992 Eagles make big jumps into the top three.

Let’s test this out on the next best team by DVOA, a very close LA Rams squad:

New weighting

Equal weighted

1989 49ers still right there amongst the top, but the 2023 49ers make a bit of an appearance on the weighted version. Z-score get smaller, but that might just be pure optics if the ranking of the teams doesn’t change much. In any case, I’ll keep playing around with weights.

Just to round out things a bit, here are the next three highest DVOA teams

Detroit Lions

Indianapolis Colts

Green Bay Packers

But really the purpose of this piece is to look at all the current playoff contenders and find historical comparisons.

To define playoff contender, let’s keep a loose criteria here and define it as any team with a 10% or higher chance of making the playoffs, per FTN (since we’ll stick with the DVOA-based theme here). You can look at other sites for their playoff odds, but keeping “in-house” at FTN on this one for odds and historical comps.

We’ll find the three closest historical comparisons (by the z-score method we’ve been using) and then how those teams did in the playoffs.

Sorted by FTN playoff odds

Right off the bat, the three closest teams to the 2025 Patriots not only didn’t win the Super Bowl, but they didn’t even make the playoffs. DVOA is very down on the Patriots, right now having them as the 14th best team by total DVOA.

And if fact if you zoom out, a lot of the closest teams didn’t make the playoffs either, or if they did, they were bounced early (note that the playoff rounds have changed over the years and at times, the divisional or wild card round could have been the first or second round depending on the year)

You’ve got to get a good bit down the list before we find a Super Bowl participant in the infamous 2016 Falcons.

Moving onto the team the next highest playoff odds, and the likely AFC West winner, the Denver Broncos.

By total DVOA they are 8th overall, certainly a strong team defensively (though not with as gawdy numbers by DVOA as the raw stats), an about average offense and special teams. Sum it all up and you get a solidly above average team, whose closest comparisons made the playoffs, but were bounced early as well.

We’ve covered the Seahawks and Rams already, two teams with very good historical comparisons that should be the NFC Championship matchup if the playoff table works out as possible.

But what I really wanted to point out again (re-posting the image to just bring it back to the front for you) is that of the 20 teams with playoff odds of >=10%, 12 of them their very closest historical comparison didn’t make the playoffs at all. This includes a few teams that a surely going to make the playoffs in New England, San Francisco, Jacksonville, Philadelphia, and Tampa Bay.

Let’s do one more team view, the Buffalo Bills. They sit with very likely playoff odds (either through winning their division race against the DVOA hated Patriots or being one of the first teams in via the wild card) and have as their comparison a team that beat a juggernaut team in the Super Bowl (Giants beating the 18-0 Patriots), the “We want the ball and we’re going to score” 2003 Seahawks that infamously did not score (Green Bay did), and then the 2006 Saints who lost in the championship game that funny enough got flack for being a dome team that can’t play well in the snow (unlike our Bills here).

The Bills aren’t loved by DVOA (they are the 10th best team overall), but at the top there they had teams with decent playoff success. A Super Bowl winner and two teams that made it to the NFC conference championship (and lost - a very current Bills thing to do).

Hopefully this was an interesting read as we get prepared for the final four games of the season. I’d be betting on (not gambling advice) the Super Bowl winner to be coming from the NFC side.

The AFC features some brutal comparisons for the top teams.

Hi. I am back for a post after an extended break. It’s not because I was on an exotic vacation or just was too busy in my life to write anything. It’s just that frankly other than one other idea (which I may still do), I didn’t have anything itching writing-wise for me to scratch.

But hey, something popped up, so here we are. Me behind my laptop typing, you pressing delete on your email notification of this new post.

I have two favorite football teams: the Chiefs and the Ravens. It’s an odd pairing, I know, but it is what it is and I’ve explained it before. Last night, one of those two teams had a miserable loss. Well, both teams had miserable losses but only one of them was really miserable (Ravens).

The Chiefs loss to the Chargers on another continent was more frustrating but well within the realm of expectations. They never had a win probability higher than ~60% and even made a little bit of a comeback before the Chargers took over for good (per rbsdm.com).

But take a look at this sucker:

SHEEEEEEEESH!

That’s a chart that shows the Ravens win probability bumping it’s head on the top of the chart almost. When the Bills punted with 9:27 left in the 4th, down 40-25, the Ravens win probability was 99.23%. If you, like me, took consumer math in high school when everyone else was in algebra II, you may or may not know that the highest a win probability figure can get to is 100% (you’ve won the game - congrats!). But there isn’t much room between 100% and 99.23%. And while functionally they are pretty much the same number (about as close as something in this world can be for certain but hasn’t happened), in reality, 99.23% =/= 100%.

So yeah, the Ravens by any meaning of the word “blew”, blew that game. And Ravens fans again were left sobbing into their fried lake trout when the angrily raided the fridge yet again after seeing another collapse from their guys in purple and black. It’s a reverse of one of my favorite quotes from Ocean’s 13:

As usual when a teams blows a lead, Twitter busts out the digital pitchforks, calls for the head of the coach, and the timeline gets filled with stats like below:

and

and

But one I want to focus on is this from Ben Solak at ESPN:

(Ben is great, not a knock on him, just using this as an example)

If you are a Ravens fan, Ravens employee, Ravens player, or in particular, the Ravens head coach, this is not a fun stat. You’d ideally like to win 100% of the games you have a win probability of 90% of winning. Now that we are on it, if I were a head coach, I’d like to win 100% of games with any win probability >0%!

But the core focus of the tweet rightfully is the point that the Ravens under Harbaugh have had some pretty bad collapses when you look at the game and outcome through the lens of win probability. There have been more games than they’d like where at some point late in the game, the broadcast zooms in on John Harbaugh looking like this:

or

It’s not a coincidence I picked two Steelers games here, and we’ll get to that in a moment. But for the above game where the Ravens lost in 13-16 at home to the Steelers in 2022, let’s go to the chart

From the entire 2nd quarter through damn near the end of the game, the Ravens had >75% probability of winning per the win models. And yet, when the clock drained down near 0:00, John Harbaugh was left looking like this

Back to Ben’s tweet, I wanted to run some numbers for myself and use a bit different criteria and add some context.

edit: after some feedback on the numbers, I re-ran the everything and recalculated. The general percentages are the same, but more games are included in the sample with the revised logic.

So here is every time a head coach’s team since 2019 had an 80% win probability (per the Vegas adjusted win probability model in nflfastR) and their W/L record.

70 total coaches make the list (I am just now wondering if there is a coach who *hasn’t* had an 80% win probability in a game during that span but sorry I already closed R and I have a rule) of games they “should” have won.

The quotation marks there are very intentional because, well, when should a team win? When they have an 80% win probability? When it’s 90%? When it’s 99.23%? You have to make the cutoff somewhere, and I went with 80%. Why? I don’t know, it seemed to be fine and if I told you a team had an 80% chance of winning before the game was even played, you might tell me that is too high for any NFL game. At 80%, the favored team has clearly done enough to outpace the pre-game odds and have the Vegas adjusted probability put them firmly in the drivers seat. 90% is a good baseline too, but it also might cut off a bit from our sample. Maybe 85% in the middle ground is good? Sure, but let’s roll with 80% here.

By definition though…20% of the time you should lose!

If some probability model spits out you have a 71.4% chance of winning, you shouldn’t be surprised when 28.6% of the time you lose if you happen to have the single roll of the dice that ends with that 28.6% part of the tail. You only get the one roll and the dice doesn’t have your name on every side.

Let’s zoom in a bit and just focus on a handful of coaches here that pass the sample size of having 20 games reaching that win probability (sorry Jerry Rosburg)

What do you think this list would look like if I gave you these coaches, the times they reached the threshold, and asked “how many times did they win?” You surely wouldn’t say they all won 100% of the time. I think you’d say guys like Andy Reid, Bill Belichick, Mike Tomlin, and Dan Campbell probably won a lot of their games. You’d probably guess that the subject of our post here lost a lot of them, same with Kyle Shanahan. I bet you probably thought Sean McDermott would be higher as he’s had some high profile collapses/losses.

But sometimes whether you win or lose, it’s just bad luck. There are complaints about Harbaugh from the game on Sunday night against the Bills. Valid ones that added some signal to the outcome. Baltimore allowed a field goal at the end of the half in part because they called a play with 2:03 left in the half rather than run it down to the two minute warning. This stopped the clock in the Bills favor and if you recall, Buffalo attempted their halftime field goal with 0:01 left on the clock. That doesn’t happen if the Ravens drain those extra three seconds off. Also on the game losing field goal, Harbaugh took a timeout at 0:34 during a Bills 2nd down kneel. The Bills were trying to drain the clock but they themselves were without any timeouts and the Ravens had one left, putting them in control of the very clock their opponent was trying to manipulate. When Harbaugh called a timeout at 0:34 left, he knew that all the bills could do was kneel and then fire drill their kicking team out with a running clock. But calling timeout with 0:34 left meant that the Bills would have ~32 seconds between the time Allen’s knee touched the ground and the field goal had to be kicked. The field goal was kicked with technically 0:03 seconds left but the Bills did take a few seconds in luxury because they wanted to drain as much as they could off the clock. But if Harbaugh calls that timeout with less time on the clock, say, 20 seconds left in the game, the Bills have a much tougher scenarios of having to kneel and fire drill. Or maybe they just kick the FG on 3rd down with a stopped clock, but that would force them to at least have to kick the ball off to the Ravens with ~15 seconds left.

The really big moment though was the inexcusable punt that both major 4th down decision modelers were screaming as a go, both with similar net WPA outcomes on for the attempt.

It’s game over if you convert, and if you don’t, Buffalo has the ball at your own 39 but with 1:33 left on the clock and all your timeouts, it’s not unreasonable to think you will have a shot with some decent time left on the clock to either kick the game winning field goal or go for the game winning touchdown depending on the Bills drive outcome (and hey, maybe you stop them!).

Aaron Schatz at FTN last year found something similar using DVOA and post-game win expectancy (PGWE).

Post-Game Win Expectancy, for those who are new to my stats, is a metric that attempts to figure out “based on the stats of this game, how often would team X have won?” It is based on the stats of the game itself, not any ratings going into the game. It incorporates VOA (without opponent adjustment) along with the number of plays each team has (more plays is better) and the number of penalties each team has (fewer penalties is better)

There is Harbaugh again, at the bottom of the list (though Kyle Shanahan isn’t included in the screenshot but was mentioned earlier in the piece and far outpaces Harbaugh with a -7 of his own, basically the inverse of Andy Reid).

But there is still some strong noise here too. Do you think it’s Harbaugh’s fault his hall of fame running back fumbled the ball on what was hoping to be the game sealing drive? A accidentally 4th down deflected pass from Allen by his own teammate ends up in a receivers hands in the end zone? That in a game vs the Giants in 2022 his hall of fame QB would throw a late interception on a bad snap, then get strip sacked a few minutes later when they get the ball back? That Gus Edwards would fumble at their own 10 yard line late against the Jaguars? In that same game after mounting a go-ahead TD with 2 minutes left, on a Trevor Lawrence strip sack the ball bounces away from two Ravens defenders right into the arms of an offensive lineman? That Mark Andrews against the Steelers would drop the game-winning 2PAT? That Justin Tucker would miss two field goals and an extra point in a game vs the Eagles in 2024?

That isn’t to say it’s all just bad or good luck.

One interesting thing here is that if we go back to the 20+ game coaches list, Mike Tomlin is 25-1 in those games, his sole loss being a game vs the Commanders in 2020 that Pittsburgh played down a few players due to COVID and injuries on 5 days rest. Going for the game winning or tying drive down three with ~2 minutes left, his hall of fame QB throws a tipped pass at the line that bounces 20 feet high and into the arms of a linebacker 10 yards away.

And there are other similar fortunate tales/tail results for Tomlin winning a game they “should” have lost, including the aforementioned drop 2PAT pass to Andrews.

If Tomlin is on the Mt Rushmore of winning games they should win, it’s in part due to the subject of this post.

Of Harbaughs fourteen losses, four of them have been against Tomlin and the only coach who appears multiple times. That’s not great when that opponent is your division rival…

But therein lies part of the point too, volume. Yes, the Ravens have lost 14 games when their win probability is 80%+, but at the same time, they’ve played in more of those games outside of the Andy Reid Chiefs. Even if you applied Andy Reid’s winning% in those games, he’d still be third on the list in total losses.

I’ll present you with a would you rather have:

Coach A: “leads” your team to a lot of games where they are strong favorites to win during the game but has let’s call it an above average number of “blown” losses both in total and rates

Coach B: A coach who doesn’t have many games where the team is strong favorites

Ideally you’d like Coach C, leads to a lot of those games and doesn’t blow any of them but those weren’t the options.

You hopefully are choosing Coach A, who in this pick-em is John Harbaugh. It can’t be solely that roster building and good luck has the Ravens with the 3rd most wins in the NFL since 2019, which includes periods of times that their star QB was hurt, had COVID, or rested in the final game because they had the #1 seed locked up.

Harbaugh was imperfect on Sunday night, absolutely, but he was also failed by his team at moments and punished by the football gods. Think back to two of the toughest games from a fan perspective in the Lamar era.

2019 Divisional Round vs TEN - Lamar had two interceptions, one that bounced off Andrews hands. He also had a fumble. The Ravens had been the best team in the league on 4th down all year but went 0-2 on crucial moments.

2023 AFC Championship Game vs KC - Zay Flowers makes a great play deep down the field and then prompts to fumble at the goal line a play later. Lamar forces a pass into triple coverage and also fumbles earlier in the game.

How much of that unfortunate moments are on Harbaugh? How much credit does he get for the team even getting to those games? Surely it’s some from both and it’s not zero for each perhaps?

And this isn’t meant to absolve Harbaugh from the disappointment. But more so to point out that disappointment is caused by high expectations that are created, and Harbaugh has done something right to set the bar that high, even if they hit their head on it.

1. I am not a Raiders fan (I support the Chiefs and Ravens - don’t worry about that). I’ve been to Las Vegas several times (never been to Oakland) and I love the town, but I don’t root for the Raider franchise or care that much what they do.

2. As it stands, the Raiders have no clear path to either of the top two consensus rookie QBs (Cam Ward and Shedeur Sanders), there isn’t some great QB available for trade that we know of, and it doesn’t appear to be shaping up that a truly difference making QB will be available in free agency either (hardly any ever are). So the Raiders biggest handicap vs many of the other head coaching and general manager openings is a real one.

The head coach and general manager openings on Black Monday were as follows:

Head Coach

Chicago Bears - filled by Ben Johnson

New England Patriots - filled by Mike Vrabel

New York Jets - Open

Jacksonville Jaguars - Open

Las Vegas Raiders - Open

Dallas Cowboys - Open

New Orleans Saints - Open

GM

Tennessee Titans - Filled by Mike Borgonzi

New York Jets - Open

Las Vegas Raiders - Open

As you can see, only two teams have both openings open. This isn’t a good or bad thing in and of itself and you could look at it through either lens. If you are a HC candidate, you could be paired with a bad GM (like in Jacksonville) or in the case of LV and NYJ, no GM and potentially be a part of that conversation for hiring one. This also means that in some way, your tenure as a head coach is tied to that decision, a staff member who can (and probably will) be a big part in deciding your fate with the franchise.

Go to an open HC job with a good GM? Great, but it might mean that you are on the chopping block first if something out of your control (injuries, the great GM having a bad stretch, etc.)

Go to an open HC job with no GM? It’s nice you could be a part of that conversation and getting to pick what is ultimately a partnership and who your boss (typically) is. But you are rolling the dice as well on the unknown-ness that is NFL executive and coaching hiring (seriously we have no idea who is and isn’t a a good HC or GM candidate mostly).

Go to an open HC job with a bad GM? Obvious issues here but perhaps you can Game of Thrones your way into power and removing said bad GM. Also the GM could be the scapegoat and presumably they are on thin ice already since the HC position is open, typically meaning that the team has been bad for 2+ seasons in a row.

As for GMs, you don’t have a choice this year. You are bringing in a new coach now that the Titans role as been filled. Congrats, you get to make a high leverage decision as the first thing you do. Good luck!

Now to the point: I think the Raiders head coaching job (and the GM role too) is being underrated. Let’s do an exercise here. I’m going to Google “NFL head coach opening ranked” and pick the first few top results from major outlets that show up (congrats to SEO staffers) and see where the Raiders job ranks.

ESPN: 5th (JAX/CHI/NE/LV/NYJ/NO) - Dallas job was not open at the time of publishing but I’m presuming it wouldn’t be lower than LV. Also Barnwell had Miami (which I think was incorrect in even including or speculating on) and the Giants on his list (retained everyone)

USA Today: 7th (CHI/DAL/JAX/NYJ/NO/LV) - NE job not included at time of publishing because it was filled but presuming it wouldn’t be lower than LV

Fox Sports: 6th (NE/DAL/JAX/CHI/NYJ/LV/NO)

SI: 4.5 (NE/CHI/JAX/LV/NJ/NO) - JAX and LV were tied for 3A and 3B. Dallas job was not open at the time of publishing but I’m presuming it wouldn’t be lower than LV

USA Today: 7th (NE/CHI/JAX/NYJ/NO/LV) - Dallas job was not open at the time of publishing but I’m presuming it wouldn’t be lower than LV

Sporting News: 5th(CHI/DAL/JAX/LV/NO/NYJ) - NE job not included at time of publishing because it was filled but presuming it wouldn’t be lower than LV

SB Nation: 6th (CHI/NE/JAX/NYJ/LV/NO) - Dallas job was not open at the time of publishing but I’m presuming it wouldn’t be lower than LV

In a few cases, the Raiders job was ranked last. In other cases, it was only second to last. At best, it was the fourth and a half best opening. And even including that 4.5, the average was 5.9 out of 7 (did some guesswork below on missing parts and ranking but it’s roughly right in my opinion).

The consensus is that the Saints job stinks, and yeah, it does, the rankers got that one right. But I think they are underrating the Raiders job. Not that I think it is a great job, but if I had to rank them I would put it as:

1. Patriots

2. Bears

3. Cowboys

4. Raiders

5. Jaguars

6. Jets

7. Saints

And if I am being honest, I wouldn’t have an issue with it being third if you wanted to swap the Cowboys or Bears with Las Vegas.

So I guess, I should lay out my pitch why at this point.

Offseason assets

I’m going to borrow heavily here from Timo Riske and the folks at PFF, who have gathered and laid out the information both numerically and visually.

From a pure draft capital standpoint, the Raiders have the 3rd most of any team. Yes, they don’t have a top five pick (6th as of now) but they make up for that in quantity

From a pure draft capital standpoint, the Raiders have the 3rd most of any team. Yes, they don’t have a top five pick (6th as of now) but they make up for that in quantity. Their current picks (per Tankathon):

(note that the Raiders have the 6th pick in the first but the 5th and 4th in the second and third due to cycling picks with New England and Jacksonville after round one)

For all intents and purposes, the Raiders have functionally five top 100 picks. Not a world breaking amount (several other teams have five as well) but plenty of potential spots to add early round talent at the natural pick or someone who “falls” (careful calling guys a draft reach or steal during the draft).

And per PFF’s chart, only the Browns (who have the #2 pick) and the Giants (who have the #3 pick) have more total draft capital than Las Vegas, and that is despite Vegas picking 3-4 spots later in the first.

I also did some compiling of my own a few weeks back before the playoffs using the Fitzgerald-Spielberger draft value table at OverTheCap and came to a similar figure.

Again, the Raiders come in with the third most capital and the third most picks.

When it comes to cap space, the Raiders find themselves at the top of the chart (okay, right-most area of the chart) as well.

Without going too deep into effective cap space and restructuring potential, the Raiders cap picture is as good as anyone’s outside of the Patriots. The Raiders have $85M in effective cap space (cap space after paying rookie and filling out a full roster), the second most in the league. If they really wanted to open up the roof, they could have $165M in cap space if they did max restructures (they won’t), per OvertheCap below.

Even with no restructures, the Raiders have a floor before any cuts or trades of ~$85M in cap space for 2025. Some simple restructures could lift that floor to ~ $141M without major future sacrifices in space.

Looking out to 2026, the Raiders have $157M in effective cap space, $10M more than the next team. That’s $40M more than the Giants and $49M more than the Patriots (two teams at the top of the draft with them).

Obviously a lot of assumptions and unknowns above, and 2025 player additions/subtractions will change 2026 cap numbers. But the point remains: the Raiders have as much financial flexibility for the next two seasons as anyone.

Combine the draft picks and cap space, a true look at offseason assets, and you can see which teams have the best means to get better. And look…it’s the Raiders and Patriots at the top once more.

The Raiders should be in conversation for every major free agent available in a March where they have a need (a lot of them!). My mantra has always been sign for need, draft for value. The Raiders can plug immediate holes (QB, RB, WR, CB) through signings, and attack the draft in April for value with their ten picks. They have about as much flexibility as any team when it comes to improving their 2025 roster and beyond.

Quarterback

Now I’ll point out again, the Patriots (Drake Maye), Jaguars (Trevor Lawrence), Cowboys (Dak Prescott), and Bears (Caleb Williams) all (seemingly) have their franchise QB, the Raiders do not. And as it stands, they likely are not going to be able to draft Ward or Sanders and there aren’t any major QBs on the trade block. They could sign someone like Sam Darnold (who was legit good in Minnesota this year but it remains to be see if his production will repeat and how much of it was Kevin O’Connell induced) or Justin Fields (a so far “busted” QB who has athletic skills but perhaps has never had a good HC/OC supporting him). There is also various darts they could throw: drafting someone like Jalen Milroe or Jaxson Dart (notice the pun?) and see what they could be, they could trade for Hendon Hooker or Joe Milton young backups with big arms type (Will Levis I suppose too).

But the biggest point remains: their path to their 2025 and beyond quarterback is unknown, and murky at best. They have options for upgrades, but they don’t have a glaring solution (beating the Jaguars and Saints in their penultimate two games assured that). Perhaps they could trade up for the right to select Ward or Sanders but they would have to both find a partner and pay the price. Both QBs floors seem to stop at #3 overall, as surely two of TEN/CLE/NYG will take one. I suppose you could talk the Browns into trading back. If they aren’t set at taking a QB (reminder that they will still be paying Watson anywhere from $35M to $60M a year in cap for a few more seasons), they are an analytical friendly team and shouldn’t mind trading back and collecting the bounty. But that of course drains a decent amount of draft capital and the Browns would surely shop the pick between Vegas and both New York teams, increasing the price. It’s possible, but for now we should assume the Raiders 2025 QB will not be Cam Ward or Shedeur Sanders.

So yes, this is a major flaw but it isn’t necessarily fatal, especially when you look at the longer game. The incoming head coach of the Raiders doesn’t HAVE to fix the QB situation immediately and the expectations for the 2025 season shouldn’t be Super Bowl or bust. The expectation should be a long runway, being able to build out a roster and find a QB at some point. The 2026 QB class looks a lot better, and depending upon the outcome of the 2025 season and assuming the Raiders don’t make a deep playoff run, they should be in striking distance for any of the top QB prospects next year.

And ultimately, this comes down to a selection for the head coach candidate: do I want to compete now with mostly the players in place, or do I want to be part of building a roster to match my style of play and coaching. There isn’t a correct answer here. We’ve seen it work in both cases and we’ve seen it fail in both cases too.

Current roster

The Raiders current roster isn’t bereft of talent and for the most part it is cost controlled and cheap. In fact, the Raiders are one of 12 teams with positive surplus value, as measured by OvertheCap’s Value Over APY metric.

Headline players obviously include tight end phenom Brock Bowers and stalwart center potential Jackson Powers-Johnson, but there are some solid contributing young(er) players in Jakobi Meyers, Delmar Glaze, Tre Tucker, Thayer Munford, Thomas Harper, Jakorian Bennett, and Amari Burney who all contributed in some form this year and have a year or more left.

They aren’t also bogged down by older players with heavy cap hits, as they have the second youngest roster in the league (again per OTC).

To be clear, this team lost 13 games in 2024. It’s not a talented roster by any means but also quarterback play brought them down overall. Even just raising the floor of performance there would go a long way before even using the cap space available and draft picks.

The elephant (condor?) in the room is what to be done with Maxx Crosby. One of the best edge rushers in the game, he turns 28 just before the season opener and has two more years left on his deal. While those two remaining years aren’t cheap (they are the 3rd and 6th highest cap hits for edge rushers in their respective years), Crosby is a defensive weapon whose contract carries a lot of value still despite the pay. It’s not unreal to think the conversations for acquiring Crosby include a first round pick.

Ownership

It doesn’t appear like this has historically been a major issue for head coaching candidates. Teams with “bad” owners (however that is measured) like the Jets, Panthers, Titans, Jaguars, Giants, Browns, and Bears haven’t really had to settle for poor candidates when hiring. In fact, the Bears just hired one of the most sought after head coach candidates in some years in Ben Johnson yesterday!

There are only 32 NFL head coaching positions in the world. These are sought after jobs with something like 5-10 candidates per opening interviewing every time. Who is signing the checks is a small factor when it comes to these sought after positions. How often is the ownership group the deciding factor? I’d argue rarely, given that the ownership group doesn’t tank the pitch to the candidate in the hiring process.

We all know Mark Davis and the embarrassing and unserious circus that has set up the tent since he inherited the franchise from his father. Not counting Hue Jackson (who Mark inherited as head coach of his now team), the Raiders have had seven head coaches since Jackon’s firing in 2011. Basically a new coach every 1.5 to 2 years. In the same span, he’s hired four general managers (not counting interim GM Champ Kelly). Stability at the Raiders executive and coaching level has been at San Andreas fault line levels, quaking when activated in a bad way.

But for prospective head coaches and GMs looking for stability, it may be coming or already here. A month ago, the league approved Davis’ sale of a 15% of stake in the franchise to a minority ownership group, just a few months after the league also approved a 10% sale to a different minority group that included notoriously Tom Brady as a part.

As of now, Davis is still the majority owner with his mother, so he’s still the lead decision maker. But there is a thought that that might change at some point soon.

Davis of course just inherited the team after his father passed, unlike many new owners who were simply rich enough to buy the team. And while no NFL team or owner is poor or really cash strapped in the current revenue climate, there are certainly “poor” owners relative to others, with Davis as one of them (though no one is shedding a tear for him). When Davis sold part of the franchise, he turned that equity ownership on paper to an actual cash flow into his and the team’s pocket. Meanwhile in taking on minority owners far wealthier than him (from a liquidity perspective), the checkbook gets a bit bigger.

In part too, Davis will have to pay inheritance tax when his mother passes away on the massive growth in value of the franchise from when it was purchased in 1966 to today. As an only child, the cost comes out of his pocket, and while it is presumed that his recent sales were in part to cover that future liability, he may sell more if necessary (and there is certainly a market as the NFL recently approved private equity groups to purchase stakes). How much future sale potentially is the question. Per league rules, the controlling owner must own at least 30% and no team can have more than 25 owners. If the Brady group and the Silver Lake/Discovery Land Company group don’t acquire any more, Davis could sell up to ~44% of remaining equity, having to split it at least between two groups (so no group has more than 30% and Davis still does). That obviously would mean Davis would give up majority control, but would still own the largest share.

(a quick note: Davis is the majority owner but he doesn’t own all of the portion that was retained after the recent sales. There are a handful of small owners as a part of his ownership block too and some of their shares were sold alongside Davis’)

This isn’t to say it will happen, but more so that Davis is no longer the sole voice running things and he’s opened up paths for others to contribute while cashing in on his financial liabilities and needs. The sole voice part is important here.

Davis brought on Brady and Co. specifically to “lean on him for organizational needs” and have a prominent voice in football operations. And Brady for his part doesn’t appear to just want to be a passive owner, in it just to say he owns a team and get a nice suite on gameday. Brady has already been involved, per The Athletic, as he’s:

• Sat in on head coaching and GM interviews

• Reached out to Bill Belichick while the ink is drying on his UNC contract still

• Was part of the decision making group to fire Telesco and Pierce

Meanwhile in another change of pace, Davis has hired a third party firm (Korn Ferry) as a consultant and search committee for filling their open roles versus when he did all the hiring and candidate pool collection mostly himself (and with the help of others he knew).

It does remain to be seen if the Raiders have flipped the switch from an unserious organization to a serious one, but they’ve made steps. Landing Ben Johnson would have been a very big step but you can’t force someone to take a job and Johnson went with arguably the best opening available (that also paid him a boat load of money to do so). But firing Telesco and Pierce (neither should have been hired in the first place but we can chalk that up to just Davis running the show pre-sale) at least provides a clean reset and path forward.

I think there is a real appeal to this job as a head coach and general manager. From a financial/asset standpoint, you’ve got a lot working for you:

1. Massive amount of cap space

2. A high quality and quantity of draft picks

3. Several good to elite players under contract

4. A new stadium and practice facility

5. An appealing city to attract talent

6. No state income tax

7. A new ownership group working toward turning the page from the past years

It’s not a guarantee the Raiders can shake the shame of their history since the passing of Al Davis, but a head coach and GM can be a part of the inflection point of turning a historic franchise around and building it from the ground up in their vision. That’s appealing, even if it is missing a core piece of the attraction in a franchise QB (or one within reach).

Well, as Erlich Bachman would say “shit has pivoted". New datahubs from existing sites, new sites altogether, and some drastic changes and turnover for some of the “legacy” sites. There have been some major shifts in the landscape since the original post, and that’s what has prompted this new post. I always intended this Substack to just be written on an as-needed basis, with the cadence being just whenever I felt like I wanted to write about something. I think there’s been enough movement that it warrants an update.

Despite all the advancements, a gap still exists for lack of a better word, the minutia. Want to watch a particular offensive tackle vs defense end matchup? Sorry, no can do. Want to see what a good run fit looks like? Maybe sometime in the future. Want to see tight coverage on a play where the receiver wasn’t targeted? Sorry, the defender should have done a little worse job so the receiver was targeted, then you could watch.

We’ve got some more information and ability since the original piece, but given that most of the public data is fantasy focused and only a few sickos have linemen and corners as part of fantasy, there is a gap there. There are of course private solutions (like PFF Ultimate being able to jump right to specific one-on-one matchups) or the occasional public update (think about Pass Block Win Rates but they are only available in small batches and at the season level).

I’m personally probably 70% stats/30% film guy right now, but that ratio is an involuntary one. The problem is that I don’t know what good pass blocking from a center looks like. Sure, people post clips on Twitter, but it’s impossible for me to apply that over 1,000 snaps on other players. Okay, let’s say I become an expert on pass blocking, then I’ve still got linebacker play, cornerback play, wide receiver route running, etc. Everyone can tell you what a good QB play is for the most part. They can probably tell you too “that’s a nice catch by the receiver.” And of course, anyone can tell you the bad. Even the most casual fan can tell you on a play where the offensive tackle gets beaten instantly or the corner blows a coverage.

But sports live in the middle of the highs and lows. Tom Brady threw 12,000 passes for his career (and many more dropbacks) but “only” 649 of those attempts were touchdowns (5.5%) and 212 were interceptions (2%). That leaves another ~92.5% of his attempts as the middle of highs and lows, great and bad. How is the average fan, or even the above average fan, supposed to identify those in-between shades?

There’s are solutions (such PFF Ultimate) but they’re private. You can sorta scrape the data, but connecting it to the film is impossible, and the data is more focused on fantasy.

With NFL Pro, you can now see every play when a player is on the field, but the filters are a bit limited (but surely will get better over time). So if I wanted to watch every pressure from Myles Garret I can, but if I wanted to watch every blown block by a particular offensive lineman, I cannot. I can view every time that offensive lineman is on the field, but not their pluses and minuses.

And there is basically no special teams data. If the minutia is a gap, then special teams is the grand canyon. I don’t even think in NFL Pro you can look at kicker/punter snaps besides just all of them (so if you want to look at Justin Tucker field goals, you have to include kickoffs and XPs too).

The goal of analytics is to better explain what is going on, to give more insights and uncover things we didn’t think we knew, couldn’t know without the data, or didn’t think we could know. It’s also to help those whose know the film understand the film. But the inverse is important too. Football is best enjoyed when you understand it, and you can’t understand it without the film side. You can learn the data side I think easily-ish. You certainly cannot learn the film side easily-ish. Even if you wanted to watch 1,000 snaps of good offensive line play…where could you do that? You couldn’t. You need to watch 1,000 snaps and just try to figure out what is good, what is bad, and what is the in-between.

PFF

We’ll start with the big one, arguably the Apple or Microsoft of the public data space. They’ve long had some of the best writers, best data folks, their data was the best, and it was easily available in a nice looking format.

That might no longer be the case, and I use “might” as a qualifier here because I’m giving them some benefit of the doubt and holding back a “probably”. As Arif Hasan covered in his long piece just a few weeks ago, there has been significant turnover at the company. Not only has some upper level staff left (including founder Neil Hornsby) but some of the more public facing staff have departed as well. This includes faces that I know I have been watching and seeing ever since I get into following the public space (people like Eric Eager, Sam and Steve, Kevin Cole, Brad Spielberger). Some left on their own accord, but it’s hard to read Arif’s piece and not think they maybe if things were different, all those folks would still be there.

As far as the site itself, if you want an update on their data offering…just go read the original piece. Seriously. Nothing has changed. It’s the same site (Premium Stats 2.0) with not a single change I can find. All the good and the bad still exist on their portal. While other sites are making improvements, PFF’s Premium Stats 2.0 has remained stagnant for 3-4 years at this point. The last update I believe was in April of 2022 and that was mostly just a UI change with some added ability to split stats for different time periods (but not multiple time periods mind you…one of my biggest complaints). The last true major update was March of 2021.

I say this with all respect for the site and the people that built it: it feels like they’ve abandoned their public product. Their Facebook groups are a ghost town, they functionally deleted their discord server, Stats 2.0 hasn’t been updated in years, and they are losing both hosts of their most public facing NFL show (Sam and Steve are leaving for 33rd Team in a few days from now as of writing this).

Maybe PFF is no longer Microsoft or Apple…maybe they are now IBM. Once the leaders of a growing industry, they are now behind and trying to play catch up.

The thing is with PFF, the public product was never the money maker. It was once the core business because it was basically all they had. But once PFF Ultimate (and now also PFF IQ) really started to take shape (thanks in part to people like the now departed Steve Palazzolo), the real money came from teams and the services to them. And that’s fine! Landscapes change and PFF skated to where the puck was going (maybe the better metaphor is threw the ball to where the receiver was going to be). But because the revenue mix changed drastically (the public product can probably just be lumped into the same bucket as their revenue from YouTube and Twitter…), the support changed too. The allure and needs of being integrated (and pretty deeply I would imagine) with all 32 NFL teams, the league office, media networks, every FBS school, and many FCS schools weigh far more than the needs of supporting the $80 a year that Joe Sixpack from Pacoima is paying them.

Premium Stats 2.0 is an $80 product, so PFF is only giving $80 worth of features. It was once the place to get data, now it’s fallen behind.

But that doesn’t mean all hope is lost. IBM found their way back at one point, and it’s reasonable that PFF can too. I’ve half-joked that the NFL should just buy PFF at this point, and I think there could be some actual benefit to that (teams already pay a lot of money to PFF annually). But that’s unlikely (more on this later) and it’s more of a fun speculative idea than a potential reality.

What PFF needs to do is one of two things:

1) Make their Stats 2.0 a better offering (PFF I have thoughts on this if someone from the company somehow reads this). This includes some real ground-moving work but also simple things like I’ve mentioned continuously like being able to look at multiple years of stats. Some of these things are just layups they are intentionally choosing not to attempt. They already have all these features available in PFF Ultimate. And many of these features are available elsewhere. There’s no excuse for Stats 2.0 to be as stagnant as it has been other than PFF just simply not caring for the product anymore and just taking the money they know they can keep getting.

2) Make a public PFF Ultimate-like product. I will give PFF a lot of credit here: there is simply no football product, public or private, that is better than Ultimate. It does everything, it’s UI is user friendly, and it is snappy. They could surely launch a slimmed down version of PFF Ultimate and charge users 5-6x what they are paying for PFF+ (let’s call it PFF Advanced). The market is there and similar companies like SIS are charging $800 for a similar public product. One of the best things about PFF Ultimate is you can jump directly to the video of the plays you have filtered for, as well as getting the charting of the play and the NGS-like dots. They won’t do this for PFF Advanced given the licensing rules (and now that NextGen/NFL have a version of this), but it’s just a point in how great Ultimate is. PFF Advanced could give many of the same features as Ultimate (minus the charting, dots, and film), but do so in a way that maintains Ultimate as their core and best product (alongside PFF IQ). They can pull forward PFF+ without pushing down PFF Ultimate. Teams aren’t going to abandon PFF Ultimate (and the money that comes with it for PFF) because PFF let public users look at turnover-worthy throw rates for multiple years instead of just on a single year basis (something you cannot do today). There is ground that can be closed between PFF+ and PFF Ultimate without taking away business from the team product, while increasing business from the public product. It’s a win-win, but PFF doesn’t seem to either realize that or care. And I want to be clear: PFF Ultimate is not exclusively a team product. It’s not like they are reserving this solely for NFL/FBS/FCS teams. Ultimate is also available to media outlets (for instance The Ringer has access to it) and sports agencies. Could they be at risk of losing some business from media/agencies as PFF Advanced cannibalizes some of those users downgrading from Ultimate to Advanced? Sure. But I would imagine the revenue would be made up and more from users upgrading from PFF+ to PFF Advanced.

If you think of the public wallet as a finite amount of money, which gets budgeted between datasites, PFF was once the majority (even the sole) of the spend. But with new products elsewhere being launched, the wallet size likely hasn’t grown, and PFF is getting a smaller piece of the pie.

And it’s worth pointing out, that team wallets have now been re-allocated. PFF is no longer alone in offering a team product. For instance, StatsBomb has been rolling out a NFL product that matches and even exceeds PFF Ultimate in some matters. Teams also have the raw NGS data of course, so they can build in-house their own player tracking and metrics instead of relying on PFF to do it (either via hand charting or their player tracking models).

NFL Pro

It’s not a coincidence that I put these two sites in order, at the start of this piece. The ground that PFF had lost up until last week to other providers was real but not a crazy amount. But as of last week, they have now lost maybe a crazy amount of ground with the launch of NFL Pro. It’s not a mortal wound that PFF has suffered, but it is a real gutshot and if they don’t stop the bleeding, it could turn fatal or at least a period of agonizing pain.

NFL Pro is now public. And it’s really great.

As mentioned earlier, one thing PFF Ultimate has that is maybe the product defining feature is the ability to do mass filtering and then instantly watch and cutup all the clips of those plays. NFL Pro now gives the public that availability.

If I wanted to see how Brock Purdy performed when blitzed in 2023 and see all those plays, I can do that right from the leaderboard.

If I wanted to see how wide receivers did in tight window throws, I simply add the filter.

You want to see how many times Marlon Humphrey was targeted in press covered? Okay. Done.

I mentioned this specific idea in my original piece because the MLB was already giving the public the ability to do this with their Statcast/Baseball Savant product. Now, we know the NFL is a bit tighter with their film rights, so while Savant is 100% free, NFL Pro does come at a price ($15 a month or $99 a year). But it’s such a small cost compared to the overall product, that it is worth it. And remember, this is the launch product. It’s only going to get better and from what I understand, the team behind it is going to work busily on adding new features, functions, and tools. I wouldn’t be surprised if NFL Pro becomes the highest share of the public user data wallet. This is the closest thing for now that the public will get to something like PFF Ultimate.

I’ve seen some people talk about how this product spells doom for PFF and I disagree. NFL Pro, as it currently stands, is a public facing tool (I haven’t seen any hint or evidence there is a team facing one available). PFF Ultimate is a team facing tool (and media/sports agencies). There is little to no overlap between the two. We know PFF Ultimate is the main revenue driver of PFF, and there will surely not be any teams that are going to rush to cancel their PFF Ultimate subscription and sign up for NFL Pro. Even the cost difference savings of what would be hundreds of thousands for PFF Ultimate vs maybe a few thousand for the entire front staff having individual NFL Pro subscriptions isn’t worth the downgrade.

As things are right now, NFL Pro will likely be the premier public facing tool since it offers something no other public tool can: video. As far as I am aware, NFL Pro is the only place you can jump right to clips from or view any All-22 film. And that makes sense, given that NFL is the rightsholder. There are a few private products that allow for film integration (PFF Ultimate, SIS Datahub, etc) but 1) you have to work out additional licenses and 2) from what I understand the NFL is pretty particular with the video you can show. There doesn’t appear to be any restrictions on the face of things with NFL Pro and video sharing (ie: you can share All-22 clips from NFL Pro).

StatsBomb

This piece is meant to really look at just public facing analytics, but I think I need to write quick blurb about StatsBomb.

They are the biggest challenger to the private side for PFF and already work with NFL and college teams (I don’t know to what capacity or numbers but I’d assume it’s most the NFL and dozens of FBS teams). Having said that, they have no public facing tool. I have no idea if they are close to launching one or will ever. StatsBomb first made a name for themselves in European soccer around 11-12 years ago and there isn’t currently even a public tool for that.

StatsBomb has released some NFL data as part of a salute to Tom Brady’s career but that’s really about it outside of their few times a week tweets on their Twitter account.

They are the leader in soccer data on the team side, and could be the leader in the NFL on the team side too. Could they be the leader on the public side? Sure, if they wanted to. It doesn’t seem like they want to though (not a knock on them, like PFF knows, the money is on the team side).

FTN StatsHub

The newest entry (sorta) into the space. I didn’t have them in my original piece, because, well, they didn’t really exist at that time. I did have Football Outsiders (RIP) listed but after a confusing year and change for the website (including ownership not honoring contracts and pay), FO is gone. The website is completely gone, as is the entire archive and history.

BUT! Football Outsiders lives on in a more concentrated capacity over at FTN (in case you wanted to know what FTN stands for it is “For the Numbers”). Most importantly, DVOA and the DVOA database found a good home and that lives on, even if all of Aaron and his crews’ great weekly pieces and recaps now live on only in our memories.

The major strength of FTN (DVOA is great still of course) is that in July, they released their StatsHub. A portal that allows you to do numerous filters, and deep one too.

Want to know which QB had the highest total EPA on throwing to post route in tight coverage vs cover four? It’s CJ Stroud.

Which running back had the best success rate on outside zone runs into stacked boxes? Add the filters.

Are you curious which receiver had the most targets on early downs on go routes with tight or one step coverage in rainy weather? You are clearly sick in the head and should seek help, but hey, who am I to judge and you can filter for it.

And FTN will let you view not just at the player level, but the often missed out team level that some providers don’t have. The filters are pretty inclusive (including having the coverage scheme available, which is really important) as well as charting data. And I’ve been told that FTN does their charting in-house rather than license it, meaning they control their own data and can improve it rather than rely on a provider. Always a benefit.

And as a bonus, since FTN is now the source for DVOA data, you can do your own filters for DVOA at various player and team levels, as well as have access to the overall DVOA database with the same subscription.

FTN StatsHub is still pretty new, so I can’t give a pound the desk type endorsement, but it’s a good product and the team at FTN is dedicated to making it a best in class solution (plus they have a good staff there from some of the people I’ve interacted with), so I’m optimistic. Also to be honest, it’s not that expensive? I guess cost is relative to budget, but as far as alternate solutions, it’s probably the best value right now.

33rd Team

I mentioned 33rd Team in the original post and not a ton on the site itself seems to have changed necessarily. The Edge is still a nice tool, even if it is limited a bit in filters and it's still geared towards fantasy.

Unfortunately it looks like they somewhat de-integrated with SIS? For instance, in a prior iteration of the site, when you clicked on a team roster, you could view advanced stats of every player on the roster or position, including some really detailed stuff you couldn’t get on a SIS Datahub subscription.

Now, that’s all gone?

The stats page seems to have been removed at the team level, you can’t click on any of the player names/photos to take you to the player page, and on the player pages it shows very simple stats (though in a nice format!).

This player page used to allow you to dive deep into a players performance with some pretty detailed metrics like __ over expected and whatnot. Now it just shows things like passer rating and passing yards per game by depth of throw? Don’t think I’ve ever even seen that stat before. What is the use case of knowing Mac Jones averaged 127 yards per game on throws of 1-15 yards? Especially when you can’t compare it to another player?

Also The Edge tool doesn’t even appear to be accessible as a button on the site anymore? You have to navigate directly to it via the URL.

But this all ties into what I want to get at: they seem to be in some sort of a transition. Obviously, hiring Sam Monson and Steve Palazzolo from PFF wasn’t done flippantly or because they needed a podcast host. And it’s doubly as interesting because 33rd Team hired both at the same time. You’d imagine Sam and Steve spoke to each other about this and the timing wasn’t a coincidence.

With SIS integration seemingly either going away or being revamped and the hiring of some talented folks like Same and Steve (and others you can find on the masthead), I wonder if what Steve (and Sam) did for in building out PFF, they are trying to do with 33rd Team? Sure, those two may have joined 33rd Team solely for a media role but that is a waste of their talents. Not that they aren’t good at the media stuff (they are great) but they are also good at the football product stuff. I know their intro presser talks solely about the Check the Mic podcast, but their bios then call out how fundamental they were to the growth of PFF.

Again, I could be very wrong here and Sam and Steve are solely in a media role capacity, but I can’t help but think they are going to be asked to do more at the site than some writing and a 3 podcasts a week. They did those exact same things at PFF but also found time on the product side still. Could they be working on SIS integration in a similar capacity as PFF and PFF Ultimate? Maybe but those dots might be hard to connect. SIS is their own thing and has been around forever.

33rd Team was founded by former NFL exec Mike Tannenbaum (he’s also the current CEO) and 33rd Team describes themselves as a football think tank. A think tank classically, well, thinks about their subject (football in this case) by hiring experts who create policy, rules, and do research for companies in the industry. That could be the Sam/Steve connection, as they use their years of knowledge gathered over a decade plus at PFF. They both certainly count as experts at the macro level of the game and how to think about it from both a team and consumer basis.

Not long ago 33rd Team and Infinite Athlete paired up. For those unfamiliar with Infinite Athlete, they are a technology company not dissimilar to StatsBomb, using technology, video, and player tracking information to quantify their respective sports. If you are a European soccer fan, you might recall the name of the company because it’s currently plastered on the front of Chelsea kits.

You can learn a bit more about them here and apparently the 49ers, NFL, and NFLPA are already connected in various capacities to IA and it’s subsidiaries (Biocore and Tempus Ex Machina).

So my purely speculatory grand plan with Same/Steve and 33rd Team is they are going to build a PFF/StatsBomb like product at 33rd Team. PFF is hand charted (though they do have access to The Dots) and StatsBomb is a mix of both both leans to being a tech company using video tracking. So likely 33rd Team is going to be closer to StatsBomb than PFF.

Will there ever be a public tool from them? Who knows, but it’s worth watching to see how things play out there.

Fantasy Points/Fantasy Points Data

Another new entry into the universe (a good thing!), Fantasy Points Data (or FPD from now on in this piece) started by fantasy guru (his nickname) John Hansen, which includes several former PFF charters/analysts as well as staffers from FTN with similar roles too. Safe to say, both FDP and FTN staff have taken what they learned in the grading and charting experience and are applying it now at their respective sites.

FDP is cool. It’s similar to FTN’s StatsHub, but a little more fantasy focused (given the name). But it also has more filters than FTN StatsHub (which on the face is good but it doesn’t mean it is better).

And interesting thing that FDP is doing with their data is some qualitative stats. They have things like accurate and off-target throws, as well as hero and turnover-worthy throws. That might sound familiar, as it’s the equivalent to a big time throw and turnover-worthy throw that PFF has been doing for years (FTN also has a similar stat). I think there is a small difference. PFF’s big time throw was just on the throw itself I believe, whereas the Hero stat describes the overall play itself. This seems closer to PFF giving the QB a +1-2 grade rather than a big time throw. And also, PFF is turnover-worthy play, not throw like at FDP. So a fumble would be included in TWP at PFF but not in TWT at FPD. They also have sack avoidance and sack fault, which is a stat PFF has tracked forever as well.

.

Much like FTN, the filters really are the killer feature of the overall product and I think it does filters better than FTN. You can arrange columns:

Set minimum values for any stat instead of just playing time filters (FTN allows this too)

And honestly probably the deepest level of charting you can get outside of PFF Ultimate. It includes different ways to dice the data, has a lot of charting based filters, can be grouped at the team level (offense or defense), and even has the ability to do splits (which I don’t think exists anywhere else).

So if you wanted to know which Patriots receivers in October through December had the most targets when lined up out wide, in an empty backfield, against cover four and six, with motion before the snap, in shotgun, when the quarterback was out of the pocket, and he left the pocket due to pressure, when the QB was on the move, it was a catchable pass, that was contested, while the receiver was being double-teamed, the defense blitzed two defenders, and the defense was running a stunt, while the team was down by 10 in the 3rd or 4th quarter…you can do it! There are no results to show because that didn’t happen, but if it did, then you could get the information.

The combinations and way to split things out really is incredible. Now easily you can see how a QB has performed by things like coverage type, without having to add/subtract a filter one at a time. You can just do it by the split feature (this is super common in baseball but never seems to have taken hold in football stats).

And best of all, it’s exportable and in custom exportable formats.

This is the closest product to PFF Ultimate, just without the video and some of the visualization features that Ultimate has. But the tradeoff there: it doesn’t cost a tens of thousands of dollars and you don’t have to be a team or media member.

Very good product, and the $200 they are asking for it is reasonable when you consider all the hands and levers it takes to run the product.

NFL Pro

Feels like I am burying the lede here a little, as the big dog has entered the kennel (workshopping this phrase still). I teased this a bit back in the original piece and it is fully out now. I mentioned earlier that I got some very gracious feedback and reaching out when I published the original piece, and someone was able to get me access to the beta (very thankful for you!)

I’ve been messing with it for a bit now pre-launch and even more when it was publicly launched because I wanted to hype it up a bit on Twitter. It’s not the complete thread that connects film nerds and data nerds, but it is a strand. For the first time, the public user can immediately jump to highlights and cutups of specific plays in All-22.

So if you want to know about every snap of Sam Howell using play action, when the defense is in dime, blitzes, and Howell is pressured, you get more than just the data, you get the film too (and the players on the field too).

Or the times the Bears defense faced shotgun, and had four linemen, press coverage, and blitzed last year.

And it’s not just a product for the film grinders either, because now users get real access to NextGenStats and leaderboards.

The little yellow boxes on some of the cells? You can click them and it jumps right to video. One thing I am excited about using more for some positions is the ability to have more insight into top speeds. If you click on the yellow cell, you can view all the runs where for example Breece Hall eclipsed 20 or more miles per hour

Individual player pages are available too

Okay, I guess this is the closest thing to PFF Ultimate because it includes the video element. But it doesn’t quite have the human charting element that PFF/FTN/FDP/SIS has, as it’s based off NextGenStats tracking data.

Big props to the NFL and NextGen teams for getting this out. I know from the film side there were a lot of calls for this video capacity, being able to slice and dice and just literally filter plays better. And obviously, the nerds were wanting more NextGen data too, with now the added benefit of being able to jump to the film (and I’m sure the tape grinders will love the data too and being able to make quick jumps).

NFL Pro is only going to get better too. This is just the day one/week one launch product. The crew in charge of the product has already gotten some feedback on Twitter that I’ve seen and most importantly, they are willing to listen and eager to evolve and enhance the product as it ages.

SumerSports

I’m somewhat surprised they didn’t make it in the original piece, but I can’t recall if they were around then or not. But any case, they deserve a mention here.

Sumer was launched by billionaire Paul Tudor Jones (who made his fortune from his eponymous hedge fund Tudor Investments) and includes current/former staffers from all walks of like (ex-PFF employees, ex-NFL employees, even a former NFL GM in Thomas Dimitroff).

Where the majority of the time in this piece I’ve been yattering about PFF Ultimate, SumerSports (the name refers to the earliest known civilization) is more closely aligned with a different product at PFF, called PFF IQ. Whereas PFF Ultimate is maybe more for the coaches and scheme side, PFF IQ is meant for the front office and decision making staff. PFF IQ is where PFF houses all their more roster analysis related metrics, including things like their WAR/WAA models, player projections, depth charts, contract data and similarities, transfer portal, and some of the draft prospect and percentile data you may have seen. It’s the obvious fit for them, they got the on field stuff quantified and qualified, you can then convert that to roster building and transactions.

Sumer is trying to crack into that space, one that really only PFF is the main player and Sumer might be the only potential disrupter. They are already plugged into several NFL teams and have an NCAA tool available for those teams. Sumer is trying to do the ultimate idea: define a player to a specific number (based off skill, cost, chemistry, etc.) and figure out how and if they can fit into your roster. But it’s not just “does this player fill a hole and keep us under the cap” but also how the specific player fits the roster. This is something similar to how the Rams scouted draft prospects in the recent draft. Not just saying “we need an edge rusher” but how that edge rusher specifically fits on the roster and compliments other players. Turns the position group into a force multiplier.

But that analysis isn’t what we are here for in this specific piece. They have a public tool! It gives your basis stats, but also some more advanced ones that look deeper at the skill the player brings to the field. Things like yards created (which adjusts a ball carriers performance based on situation and pre-snap look) and personnel adjusted yards per route run (which adjusts Y/RR based on the difference between actual and expected yards per route run accounting for personnel, route, distance, and down). Also they provide some nice defensive stats that include things like success rate and EPA when a player is on/off the field .

They still feel a bit quiet overall, but it’s not some major pod shop looking to take the space by storm. They are building out their product, hiring good analysts, and navigating a space that hasn’t been navigated deeply.

ESPN

This is sorta a general one, but ESPN has recently launched an analytics hub not just for football projects, but other sports. Still, you’ve got their 4th down decision tool as well as receiver tracking metrics (RTM) based off their partnership with the NFL for GPS data.

I really like RTM because it’s easy to access and manipulate, and it is something you cannot get anywhere else.

Also it passes the sniff test in that generally the receivers that RTM says are good are also usually thought of being good too.

Joseph Hefner

I’m going to give a shoutout to a local guy and online friend of mine, who has a few Shiny apps he created for things like draft trade calculator based off several models, formation data, pillbox charts for snaps, and play by play data.

Final note…

That’s it. I think we are in pretty good shape right now as public consumers, with several impressive tools recently launched of various forms. Certainly take a step forward from where we were 18 months ago when I made the original post.

I’m hopeful that things will keep getting better, as the thirst for public facing analytics is never quenched.

If I missed a site or something you think should have been covered and want to let others know, reach out to me on Twitter, replying in the comments, or emailing me.

What I am good at instead is forced marching myself into data collection through sheer will and free time, compiling either manually or automatically a broad aggregation of stats. That’s both the easy part (it’s all already there to be grabbed) and the hard part (it takes time and effort to grab).

The idea of naming it Scouts + Stats was to try to blend both the film side and the stat side. In this day of sports, regardless of the individual sport itself, we live in an era where stats cannot live without film, and film without stats. Much like the impending AI revolution to be, it’s not going to be a simple as “AI will replace (job category).” Instead, it will be (job category) that works with AI and (job category) that doesn’t. Lawyers that use AI, lawyers that don’t. Financial analysts that use AI and big data, financial analysts that don’t. Teams that use big data, teams that don’t. The camps that are in the inclusive category will be the winners.

But you cannot build a draft model solely on data, because in many cases, data lives without context and outside the physical world. And in things like sports, the physical world matters. So building a draft model without including the film and human evaluation aspect of draft prospects is to be only half-conscious and willfully blind.

But one person (outside of perhaps Dane Brugler of The Athletic) cannot watch every snap of every draft eligible prospect and come to some consensus ranking. And I’m not sure you’d want that to be your sole input too. A single person has their own biases and flaws. But that isn’t to say the data itself, though objective, is flawless or solely reliable. A merging of the two is appropriate.

So what I’ve done is gathered wide ranging data points that include for every prospect listed (273 players) various things such as position, age, consensus big board ranking, PFF stable metrics based off PFF grades, and a few other stats and info I was able to get access to.

One thing I think is important for calibration is the consensus draft rankings. History says the best singular evaluator for picking players in the draft is staying close to the consensus big board. NFL teams are generally pretty good at putting players into their correct tiers and range. For instance, NFL teams and scouts are generally good at figuring out who the best handful of receivers are in the draft, but they can’t always pick the right order. Henry Ruggs, Jerry Jeudy, CeeDee Lamb, Jalen Raegor, Justin Jefferson, Brandon Aiyuk, Michael Pittman, and Tee Higgins all went within ~20 picks of each other. Looking back, the order was wrong, even if the “right” receivers were accurately grouped. In hindsight, Justin Jefferson should have been the first receiver taken (heck he probably should have been the 2nd overall pick…), but he was the 5th. Still, he was a first rounder and a consensus top 20 prospect. NFL teams were right overall, but wrong around the edges.

And I think that’s where a draft model for someone without any skin in the game (ie: not working for a team) should try to focus. Not trying to be bold and say “this guy that is 220th on the consensus board should be the first overall pick!” but instead, trying to move players up or down from their consensus ranking based on factors beyond purely just scouting only evaluations.

Which is why I think the consensus big board should work as a bit of a brake pedal for a model, being the central feedback for saying “hey, you might be wrong here…”. That isn’t to say my draft model doesn’t have some bold takes. But instead, what I think is best is to use the model as a way to try to re-order players on the consensus board, to try to put a Justin Jefferson closer to the 1st WR taken rather than the 5th. But that isn’t to say the draft model is imperfect, and in fact, you should be weary of the model if it says that the 220th player should be taken 1st overall and that it might have that player wrong.

One other note that I’ve gotten as feedback and validated with my model, QBs are hard to not only rank on a big board, but also model as well. Steve Palazzolo at PFF has built a draft model that has a pretty good hit rate based off the deep pool of information he has at his disposal, and even he says his own model struggles with quarterbacks. There are a dozen reasons for this that you can probably think of and won’t be discussing here (though I will include quarterbacks just for completion purposes).

Okay, all that said, here are the top 80 players according to the Scouts + Stats model (I’ll go position by position in a bit):

Joe Alt is the best player in the class by the model, but it isn’t adjusted for positional value intentionally. I’m not advocating for taking Joe Alt at first overall (or any non-QB really). But instead, by the model, Alt is highly deserving of his consensus big board rank of 6th overall. He’s maybe even slightly underrated by the smallest amount.

The model is very high on two players: Quinyon Mitchell from Toledo and Laiatu Latu from UCLA. Both were incredibly productive in college and in predictive stats, rank in the top 20 of the consensus board (remember we use this as a calibrator), and have been dinged for respective issues. Mitchell’s explanation is easy: not only did he not go to a big school, it wasn’t even a power five school. If he had gone to, say, Clemson, I think he’d be discussed as the best defender in this class. Latu’s ding is a little more complicated. While it’s easy to say he is simply a medical issue, it’s a bit deeper than that and has a long backstory. But it’s worth noting he has been completely healthy the past two seasons at UCLA.

One other note I think worth mentioning here is Caleb Williams at 9th, with Drake Maye above him at 3rd. There are teams that have Maye over Williams, and Nate Tice at Yahoo/The Athletic has it in that order too. Both are top five prospects on the consensus board, so really it’s Williams’ rank at 9th (seven spots below his big board rank) that is a little more interesting than Maye at 3rd (equal to his big board rank).

PFF highlighted some of his issues in a recent article:

Obviously, Williams excels as an overall thrower in most situations, but one thing he’s struggled at in his career is avoiding negative plays and sacks. Compare that to Drake Maye, who is about as good as Williams has been (if not better in certain situations) but is a little better at avoiding sacks, and a lot better at avoiding negatives.

Some other standout model vs board players:

Higher on:

• CB Kris Abrams-Draine (Missouri) 49th model rank vs 87th big board

• CB D.J. James (Auburn) 70th model rank vs 108th big board

• DT Leonard Taylor III (Miami) 73rd model rank vs 109th big board

• TE Ben Sinnott (Kansas State) 65th model rank vs 96th big board

• DT T'Vondre Sweat (Texas) 35th model ranks vs 63 big board

• ED Gabriel Murphy (UCLA) 77th model rank vs 104th big board

• G Cooper Beebe (KSU) 29th model rank vs 56th big board

• ED Bralen Trice (Washington) 40th model rank vs 64th big board

Lower on:

• OT Troy Fautanu (Washington) 69th model rank vs 19th big board

• ED Darius Robinson (Missouri) 66th model rank vs 33rd big board

• OT Kingsley Suamataia (Yale) 76th model rank vs 43rd big board

• CB Ennis Rakestraw Jr. (Missouri) 68th model rank vs 39th big board

• OT Tyler Guyton (OU) 57th model rank vs 30th big board

• WR Roman Wilson (Michigan) 71st model rank vs 50th big board

• OT Jordan Morgan (Arizona) 55th model rank vs 35th big board

• OT Olumuyiwa Fashanu (PSU) 32nd model rank vs 13th big board

Positional Boards

Quarterbacks