The scoreboard can say 24–10 while the game still feels unsettled. Or it can say 17–14, with one team holding the ball and barely any time left, and the ending can feel almost decided.

Win probability helps explain that difference. It estimates each team’s chance of winning from the situation at that moment: the score, the time remaining, who has the ball, where they have it, and what down it is. Some models also account for timeouts and the pregame spread. The details vary by model, so two broadcasts can show different numbers for the same game.

It is a forecast, not a verdict.

Read more

Six years ago, with Alex Ovechkin closing in on his 700th career goal, the question wasn’t just when he would get there. It was how long hockey might have to wait for someone else to join him.

A lot has changed since then. Ovechkin kept scoring, a new generation of stars emerged and several of the players who once looked like the best candidates to reach 700 have taken very different paths. So, with another NHL season underway, it is worth revisiting the question with the same method I used in 2020: Who has the best chance to become the next member of the 700-goal club?

Read more

There is a difference between having a large playbook and being unpredictable.

An NFL offense can call runs in every direction, throw to every level of the field and sprinkle in quarterback runs, scrambles and sneaks. But if those calls consistently arrive in the same situations — or if one play makes the next one easier to anticipate — a defense may still have a pretty good idea of what is coming.

That raises a question I’ve wanted to quantify: Which NFL coaches are the hardest to predict?

My approach was inspired by two previous analyses of NFL play calling. Chris Coxen used Shannon entropy to measure the uncertainty created by a team’s distribution of different play types. Brian Burke approached predictability through game theory, arguing that an effective mix of plays isn’t enough: calls also need to be unpredictable within a given situation and independent of what was called previously.

I combined those ideas. First, I measured how much uncertainty each coach creates with his overall mix of play calls. Then I asked the more useful questions for a defense: How much uncertainty remains once you know the down and distance? And does knowing the previous play make the next one easier to anticipate?

Using NFL play-by-play data from 2025 through Week 4 of 2026, I measured that for each of the league’s 32 current head coaches. The results reveal some of the league’s most unpredictable play callers and some offenses that give away considerably more than you might expect.

Read more

Week 2 finished 2-2, but the record undersells how well the bets held up. Both losses came after the starting quarterback was knocked out very early, fundamentally changing the game before the handicap had much chance to play out. Injuries are part of football and the losses still count but with the other two bets cashing there’s plenty of reason for optimism heading into Week 3.

Carolina was a winner as the Panthers justified their short-road-favorite price against Atlanta. Amon-Ra St. Brown also delivered, finding the end zone to cash his anytime touchdown prop at plus money. Giants quarterback Jaxson Dart was injured on the opening drive, forcing New York to play essentially the entire game without its starting quarterback and Chicago suffered a similar fate, with the Bears losing their starting quarterback very early and changing the complexion of a bet built around their offense creating separation.

On to Week 3.

Read more

Every September, the NFL hot-seat lists begin appearing. One coach is “under pressure.” Another is “on thin ice.” Someone else has a seat that is “warming,” “simmering” or already “scorching.” Two weeks into this season, Todd Bowles, Jim Harbaugh, Dan Quinn, DeMeco Ryans, Shane Steichen and several others have already appeared on one version or another.

There is just one problem: Nobody really knows what any of those descriptions mean. Is a coach on the hot seat if there is a 10 percent chance he is gone after the season? Thirty percent? Fifty? And how much should two bad games matter compared with everything that came before them such as his tenure, previous seasons, playoff success, preseason expectations or the difficulty of the schedule?

So I tried to measure it. I built a Hot Seat Index using NFL coaching history to estimate the probability that each current head coach will not return to the same job next season. Instead of asking whether a coach’s seat feels warm, the model asks a more specific question: Historically, what happened to coaches who were in similar situations at this point in the season?

That gives us a way to track the NFL’s coaching carousel before it starts spinning and, more importantly, to see whose situation is actually changing as the season unfolds.

Read more

After signing a one-year contract this summer, Alex Ovechkin, the NHL’s all-time leading goal scorer said his focus was proving he can still help the Capitals win a Stanley Cup. Given what he’s accomplished over the past five seasons, that confidence hardly seems misplaced. Ovechkin scored 32 goals at 40 years old this past season. One year earlier, he scored 44 at 39. At 36, he scored 50. Nearly two decades after entering the NHL, he remains one of hockey’s most dangerous scorers despite averaging fewer than 18 minutes per game.

The longer Ovechkin keeps doing this, the easier it becomes to believe something fundamental has changed. Players who once would have been expected to fade into retirement are now remaining productive deep into their late 30s and even their 40s. Ovechkin has become the latest example of a growing belief that elite athletes simply don’t age the way they used to.

But hockey history suggests something very different may be happening.

Read: The math behind a generational NHL player

To see how unusual Ovechkin really is, I analyzed every NHL player who recorded at least one 30-goal season since 1980, tracking how long elite scorers remained productive, which skills disappeared first and why teams keep convincing themselves the next aging superstar will be the exception.

Read more

Week 1 gave us our first glimpse of what every NFL team might be this season. It also gave us plenty of reasons to be skeptical. One game can produce misleading final scores, unsustainable efficiency and performances that look much better, or worse, than the underlying play suggests.

For Week 2, the goal is to separate what appears repeatable from what might have been opening-week noise, then see whether the betting market has adjusted too far, not far enough or just about right. Here are three games that stand out.

Read more

For years, the NFL’s offensive evolution seemed to have a clear direction: spread the field, put the quarterback in shotgun and make defenses defend every inch of grass.

Week 1 offered a reminder that evolution doesn’t always move in a straight line.

The clues were there all summer. In July, Aaron Rodgers explained why Pittsburgh wanted to throw more often from under center on early downs.

“We gotta be under center,” Rodgers said. “We gotta be under center and get the play-action game going and the movement game going.”

The New York Giants were thinking similarly. They spent the offseason drilling Jaxson Dart on under-center play action, a style Chris Simms called perhaps “the best way to create explosive plays on the field right now.”

Then came Week 1, when one of the most noticeable tactical changes across the NFL happened before the ball was even snapped. Quarterbacks were getting out from behind center and offenses were selling the threat of the run in ways that have become less common in a shotgun-heavy league.

So I went through a decade of NFL play-by-play data to answer two questions: Was opening weekend actually the beginning of a league-wide shift? And were Rodgers and Simms onto something about why teams should make it?

Read more

The Kansas City Chiefs reached seven straight AFC championship games, played in five Super Bowls and won three. Then came 2025: Patrick Mahomes suffered a season-ending knee injury, the Chiefs finished 6-11 and missed the playoffs for the first time in a decade.

Now Mahomes is racing to be ready for Week 1 while Kansas City retools around him, adding running back Kenneth Walker III and injecting youth into a defense built around Chris Jones.

Every contender eventually reaches the point when a bad season might signal something more. So how long does an NFL championship window actually stay open? And after a decade of contention, are the Chiefs approaching the end of theirs?

Read more

There is a particular kind of anxiety that comes with watching your favorite NFL team protect a late lead. They’re up two scores, the clock is dwindling and the win-probability graphic says 97 percent, but the announcers are still mapping out the increasingly improbable sequence of events that could turn the game around. A touchdown here, an onside kick there, maybe one more possession. You know your team is probably going to win. What you really want to know is when you can stop worrying about it.

That raises a deceptively simple question: How big does an NFL lead have to be, and how little time has to remain, before we can reasonably consider the game over? Not literally over, of course, but effectively over. For this exercise, I decided to draw that line at a 99 percent chance of winning. Once a team reaches that point, a comeback is still possible, but it should be the sort of thing we remember precisely because it almost never happens.

There is already a wonderfully simple version of this idea in college basketball. Erik Haslam of Haslametrics calls it a game going “analytically final,” a rule of thumb for determining when a lead has become safe. Take the size of the lead, subtract five and square the result. If there are fewer seconds remaining than that number, the game has become analytically final.

For example, a 15-point lead in college basketball becomes analytically final with fewer than 100 seconds remaining: (15 − 5)² = 100. The beauty of the rule is that it requires nothing more than the score, the clock and a little mental arithmetic. No computer model or win-probability graphic is necessary.

Football is obviously different. Scoring comes largely in chunks of three and seven, possessions consume far more time, and the difference between leading by six, seven or eight points can fundamentally change what the trailing team needs to accomplish. But the underlying question is the same: Could NFL history reveal a similarly simple relationship between the size of a lead and the amount of time required to make that lead safe?

I analyzed more than two decades of NFL play-by-play to find out. And, somewhat remarkably, the answer turned out to be simple enough to calculate while sitting on your couch.

I started with NFL play-by-play data from 2002 through 2025, covering more than 6,000 games. At regular intervals during regulation, I recorded three things: the size of the lead, the amount of time remaining and whether the team that was ahead eventually won. Ties and overtime were excluded. The play-by-play came from nflverse, whose data extends back to 1999.

The goal wasn’t to build the most sophisticated win-probability model possible. In fact, I intentionally stripped away information that we know matters. The model doesn’t know which team has the ball, how many timeouts either side has, where the ball is on the field, who is playing quarterback or what the betting market thought of the teams before kickoff. It knows only two things you can see immediately on the television screen: the lead and the time remaining.

That distinction matters because there are already NFL win-probability models designed to answer the more complicated question. nflfastR, for example, includes its own win-probability models. I was after something different: a rule simple enough to calculate while watching a game, without pulling out a computer or consulting a win-probability graphic.

To make sure I wasn’t simply finding a formula that described games it had already seen, I divided the history into two groups. Games from 2002 through 2020 were used to establish the relationship between the lead, the clock and the eventual winner. The 2021 through 2025 seasons were then set aside as an untouched test. Whatever rule emerged from the older games would ultimately have to prove itself against those newer ones.

The first task, then, was to find the combinations of lead and time remaining at which the team in front had a 99 percent chance of winning. Once I plotted those points, something unexpected happened: what looked like it should require a complicated football model began to resemble an extraordinarily simple curve.

What does 99 percent actually look like?

The first result was a map of when different NFL leads crossed the 99 percent threshold. Some of the results fit our intuition. A 12-point lead isn’t truly safe until the final seconds, while a three-touchdown lead reaches 99 percent with more than 13 minutes remaining. Get into blowout territory and the point of no return arrives much earlier: A 28-point lead crosses the threshold with more than 29 minutes left.

The curve produced by the model can be approximated with a surprisingly simple formula:

That’s it. Take the size of the lead, subtract seven, square the result and multiply by four. The answer is the number of seconds that can remain before the lead becomes analytically final, our shorthand for roughly a 99 percent chance of winning.

Say a team leads by 21 points. Subtract seven and you get 14. Square 14 to get 196, then multiply by four to get 784 seconds, or 13 minutes and four seconds. According to the simplified rule, a 21-point lead becomes analytically final at roughly 13:04 remaining.

You may notice that’s slightly different from the 13:33 shown in the table above. The earlier number comes directly from the statistical model; this formula is a simplified version of the model’s 99 percent curve. It sacrifices a few seconds of precision in exchange for something far more useful: a rule simple enough to calculate in your head. At a 21-point lead, that approximation is off by just 29 seconds.

There is one important limitation: the formula is intended for leads of eight points or more. Below that, the math no longer has a sensible football interpretation.

A neat formula isn’t particularly useful if it only describes the data that produced it. The real test was whether it would work on games it had never seen.

Remember, I used games from 2002 through 2020 to establish the relationship between lead, time remaining and the likelihood of winning. The 2021 through 2025 seasons were held out of that process entirely. I then went back to those five seasons and asked a straightforward question: When a game crossed the threshold defined by 4 × (lead − 7)², did the team in front actually win?

There were 736 games that reached the analytically-final mark. The leader won 730 of them, a 99.2 percent win rate.

That’s about as close as you could reasonably hope to get from a rule designed to identify the point at which a team has roughly a 99 percent chance of winning. And when I applied the same rule across the entire 2002–2025 sample, the result was 3,549 wins and just 20 losses, a 99.4 percent win rate.

Of course, 99 percent isn’t 100 percent. Six times from 2021 through 2025, a team crossed the analytically-final threshold and still managed to lose.

The most extreme came in December 2022, when Indianapolis led Minnesota by 30 points with 34:48 remaining. The Colts eventually stretched the lead to 33 before the Vikings began the largest comeback in NFL history, winning 39-36 in overtime.

That wasn’t even the only spectacular collapse of 2022. Baltimore led Miami by 21 points with 13 minutes remaining, enough to cross the threshold, before the Dolphins scored four fourth-quarter touchdowns and won 42–38.

Those aren’t ordinary losses, which is precisely the point. A rule defining a 99 percent threshold should occasionally fail. And when this one does, it tends to produce exactly the sort of game we remember because what happened seemed almost impossible while we were watching it.

There are plenty of things the NFL analytically-final formula doesn’t know. It doesn’t know who has possession, how many timeouts are left, where the ball is on the field or what the down and distance is. It doesn’t know whether Patrick Mahomes or a backup quarterback is trying to protect the lead. It doesn’t know which team was favored before the game or whether one team is substantially better than the other.

A sophisticated win-probability model should account for those things, and this formula isn’t intended to replace one. It’s answering a much simpler question: Given nothing but the scoreboard and the clock, when is an NFL game effectively finished?

So the next time your team is protecting a late lead, you don’t need a win-probability graphic. Subtract seven from the lead, square the result and multiply by four.

If you’ve procrastinated on preparing for your fantasy football draft, congratulations! You’ve finally found a strategy that rewards your commitment to doing nothing.

You don’t need to memorize 200 player rankings, study every preseason depth chart or spend the next week listening to fantasy football podcasts to have a good draft. You mostly need to know when to take each position and a handful of players worth targeting when your pick arrives.

That’s the idea behind the lazy person’s guide. Using current 12-team PPR average draft position, we’ve reduced the draft to a round-by-round plan: whom to target, when to grab a running back or quarterback, when you can afford to wait and when it’s time to stop worrying about a player’s projected role and start swinging for upside.

Keep this open during your draft and let everyone else do the homework.

Average draft position based on Fantasy Football Calculator data.

Read more

Drafting the best player available sounds simple. The problem is that fantasy football drafts rarely cooperate. Players disappear earlier than expected, positional runs reshape the board and every selection changes what might be available next. The Perfect Draft accounts for that uncertainty, using expert rankings and thousands of simulated picks to build an ideal roster from every spot in a 12-team league.

I began with FantasyPros consensus rankings to establish player quality, then used 12-team, point-per-reception average draft position from Fantasy Football Calculator to model realistic availability. We simulated hundreds of drafts from each of the 12 draft slots, prioritizing starting-caliber running backs and receivers through seven rounds while allowing value at quarterback and tight end. After Round 7, we emphasized upside using FantasyPros’ best, average and worst rankings. Kicker and defense were reserved for the final two rounds.

Keep in mind: This is the perfect draft based on our 2026 outlook, which may differ sharply from public average draft positions. That’s by design. When a player goes earlier or later than expected, it can give you an advantage. Use it.

(The draft grids below were crafted for a 12-team, point-per-reception league using the following starting lineup: one quarterback, two running backs, two wide receivers, one tight end, one flex player (RB/WR/TE), one defense, a kicker and seven bench players.)

Read more

There is broad agreement about the Washington Commanders entering the 2026 NFL season: They probably won’t be very good. Beyond that, good luck finding a consensus.

USA Today and Yahoo Sports rank Washington 18th in the NFL heading toward Week 1. ESPN’s Football Power Index has the Commanders 21st. The Athletic puts them 23rd. Pro Football Talk and Fox Sports have them 26th. NFL.com’s ranking is the harshest of the bunch at 29th.

That’s an 11-place range for the same team, evaluated at essentially the same moment, before any of the predictions can be tested against an actual regular-season game.

It would be easy to look at that disagreement and conclude that somebody must be wrong. By January, we’ll know who was closest. We’ll have 17 games of evidence, a final record and a neat 1-through-32 ordering with which to grade everybody’s preseason predictions.

There’s just one problem: The final standings aren’t necessarily a reliable answer key.

An NFL season is short enough and individual games uncertain enough, that even a forecaster who somehow knew exactly how good every team was before Week 1 would routinely get parts of the final standings wrong.

So before deciding whether the Commanders should be 18th, 21st, 26th or 29th, there’s a more fundamental question worth answering:

How well can anyone actually predict the NFL standings?

Read more

By the time a fantasy football draft reaches the middle rounds, the easy part is over. The obvious starters are gone and everyone is hunting for the same thing: the bench player who turns into someone you have to start.

But what should we actually be looking for? Youth? Draft pedigree? A change of scenery? A rookie buried behind an established starter? Fantasy managers have plenty of theories about what makes a good late-round pick but those theories aren’t necessarily supported by what has happened on the field.

To find out which traits have actually mattered, I analyzed more than 1,000 running backs and wide receivers drafted in the later rounds from 2010 through 2025, then tracked which ones developed into top-24 fantasy performers. I also tested the most promising patterns against later seasons rather than simply relying on what looked good in the historical data.

Most late-round picks, unsurprisingly, went nowhere. And several of the traits fantasy managers commonly associate with upside weren’t particularly helpful. But a few characteristics meaningfully improved the odds of finding a starter and they point to a small group of players worth targeting in 2026. Average draft position based on Fantasy Football Calculator data between August 10, 2026 and August 17, 2026.

Read more

Fantasy football draft season has become a race to find this year’s league winner. Every summer brings a new wave of sleeper picks, breakout candidates and late-round values, each making the case for the player who could swing a fantasy championship.

That got me wondering whether we were asking the right question.

Rather than trying to predict the next breakout, I wanted to look backward. I went through ESPN’s league winners from the past three seasons to see what they had in common. A few patterns emerged immediately. Others ran counter to ideas that have become accepted as fantasy wisdom.

The exercise changed the way I’ll approach my 2026 drafts. I think it might change yours, too.

Read more

One of the first rules of fantasy football is that you should never reach for a player. Let the draft come to you. Take the value that’s available. If someone else wants to draft a player a round early, let them make the mistake.

It’s good advice but it’s also incomplete. Every fantasy manager has players they value more highly than the market. The difference is that most of us don’t have a disciplined way of deciding when it’s worth paying a premium. Reaching because you “have a good feeling” is a great way to light draft capital on fire. Reaching because you believe the market has materially underestimated a player’s ceiling is a much different proposition.

That’s what I wanted to identify. FantasyPros publishes consensus rankings from dozens of experts but the average ranking only tells part of the story. Buried underneath that consensus are analysts who are willing to make much more aggressive bets. If one expert ranks a player 19th overall while the market is drafting him in the sixth round, that doesn’t mean the expert is right. It does mean someone whose job is evaluating fantasy players believes that player’s realistic ceiling is dramatically higher than his current cost.

I built a simple Conviction Score to measure those opportunities. The score combines two ideas: how many rounds later the market is drafting a player than the most optimistic expert would, and how large the gap is between that expert’s ranking and the overall consensus. I’m not trying to predict league winners or prove that these players will outperform ADP. I’m looking for players where respected analysts see an outcome the rest of the market isn’t fully pricing in.

That distinction matters. A player doesn’t have to finish as the overall WR1 or RB1 to justify reaching for him. If you can draft a sixth-round receiver who produces like a second-round pick or a fifth-round tight end who gives you an elite positional advantage, you’ve created one of the biggest edges available in fantasy football. Those are the bets I’m interested in making, and they’re the players you’ll find on this list.

Read more

Every NFL season produces a handful of surprise contenders that seem to come out of nowhere. Just as reliably, a few of those teams come crashing back to earth the following year. The challenge isn’t identifying who had a great season—it’s figuring out which records were built on a foundation that is unlikely to hold up. Looking only at last year’s standings can be misleading. Looking underneath them can tell a very different story.

Rather than relying on gut instinct, I started with history. Using every NFL season from 2015 through 2024, I tested which team-level statistics actually predicted a decline in wins the following year. I looked at whether teams won more games than their point differential suggested they should have, how they performed in one-score games, whether turnover margin carried predictive value after accounting for overall team quality and whether fumble recovery rates contained any meaningful signal. Some popular regression indicators held up, others largely disappeared once the data were put to the test.

The results produced a much narrower list than you might expect. A few obvious candidates survived every filter while others with gaudy records turned out to have underlying profiles that were surprisingly sustainable. And a couple of teams that barely finished above .500 showed far more warning signs than several division winners.

Below are the teams that history says are most likely to come back to the pack in 2026 and the data behind each case.

Read more

MLB’s senior data architect, Tom Tango, posed a simple question on X after Baseball Savant rolled out its new bat-tracking swing timing metrics:

“How much do hitters sit on fastballs, and how much are they early on the other pitches?”

It caught my attention because baseball has always assumed it knows the answer. We hear hitters are “sitting dead red.” We hear pitchers trying to disrupt timing by changing speeds. But despite decades of Statcast research into velocity, tunneling and pitch design, we’ve never actually been able to measure whether hitters are timing their swings for the fastball.

The new swing timing data changes that. For the first time, Statcast classifies every swing as early, on time or late based on whether the barrel reaches the ideal contact point within seven milliseconds of perfect timing. That gives us a direct way to test one of baseball's oldest assumptions.

So I decided to see what the data says. Before looking at individual pitchers like Paul Skenes, I wanted to answer Tango’s question as simply as possible: How much do major league hitters actually sit on the fastball?

Across more than 200,000 tracked swings in 2026, hitters were late on fastballs 23 percent of the time and early just three percent. Against secondary pitches, the pattern flipped almost perfectly: hitters were early 43 percent of the time and late only five percent.

Breaking the data down by pitch type shows a clear progression. Breaking the data down by pitch type shows the same pattern. Four-seam fastballs and sinkers produce the highest rates of late swings, while sliders, changeups, sweepers and curveballs produce progressively higher rates of early swings.

The league-wide numbers are exactly what we’d expect if hitters generally gear their swings for fastball velocity. The interesting question is whether some pitchers are better than others at exploiting that tendency. However, the league averages establish a baseline but they don’t tell us whether certain pitchers are especially good at disrupting timing.

To find out, I compared every pitcher’s individual offerings to the league average for that same pitch type. In other words, Paul Skenes' slider isn't compared to his fastball, it's compared to every other slider in baseball. The same is true for every four-seam fastball, changeup, curveball and splitter.

That allows us to ask a more specific question: Which pitches make hitters significantly earlier or later than they are against the average version of that pitch?

Skenes provides an interesting first test case. With a fastball that averages nearly 99 mph and one of baseball’s best strikeout rates, he would seem like an obvious candidate to produce unusual swing timing. But when we compare his pitches to league averages, the results are more nuanced. Some of his pitches behave almost exactly like the league. Others stand out dramatically.

Skenes doesn't consistently produce unusual swing timing across his entire arsenal. His four-seam fastball, sinker and sweeper all generate early swing rates that are close to league average for those pitch types.

Two pitches stand apart. Hitters are 16 percentage points more likely to be early against Skenes' slider than against the average slider. Against his splitter, they're 32 percentage points less likely to be early than against the average splitter. Those two pitches, not his fastball, are what make Skenes' swing timing profile unusual.

Tango’s question turns out to have a straightforward answer. Across Major League Baseball, the data strongly suggest hitters are generally geared for fastball velocity. When they’re wrong, they’re usually late against velocity and early against everything else.

The broader takeaway is that the league-wide pattern isn't uniform. Comparing pitchers to the average version of each pitch shows that some offerings consistently disrupt hitters' timing more than others. In Skenes' case, the biggest outliers aren't his blazing fastball, they're his slider and splitter.

Just one year after Cal Raleigh produced arguably the greatest offensive season ever by a catcher — 60 home runs, an OPS over 1.000 and an MVP runner-up finish — the Seattle Mariners’ franchise cornerstone has become one of baseball’s least productive regulars. His batting average has fallen below .200, his power has disappeared and his quality of contact has cratered.

The obvious explanations only go so far. Raleigh battled an oblique injury, and the World Baseball Classic disrupted his spring. Both probably contributed, yet neither fully explains why one of baseball’s most dangerous hitters suddenly stopped driving the ball.

The answer is hidden in a change measured in just a few inches.

Read more

Fantasy football drafts always begin with the same question: Which players should we target?

But before debating sleepers, busts and breakout candidates, it helps to understand what the market is already telling us.

Early 2026 average draft position from Fantasy Football Calculator data reveals a clear shift in how fantasy managers are approaching this season. Elite running backs are back at the top of drafts but only if they’re young. Bijan Robinson and Jahmyr Gibbs are the first two players off the board, while Ashton Jeanty and Omarion Hampton have already joined the group of franchise fantasy backs.

Wide receiver tells a different story. The elite names remain valuable, but the position is deep enough that managers don’t need to force an early pick. Tight end, meanwhile, has become expensive again. Quarterback hasn’t.

Taken together, the market is sketching out an early blueprint for 2026: prioritize scarce young talent, trust the depth at wide receiver and resist the urge to solve every position in the opening rounds.

Read more