If the Dallas Mavericks' infamous trade of Luka Doncic produced waves of shock and outrage across the NBA, then the Boston Celtics' trade of Jaylen Brown a year later was met with considerably more complicated reactions. Outrage, yes, from Celtics fans -- but also a fierce debate about Brown's on-court value and broader questions about how to analyze players in the modern, data-infused NBA.

Brown is a five-time All-Star, a former Finals MVP and the sixth-place finisher in regular-season MVP voting in 2026, after he averaged a career-best 28.7 points per game. Yet, advanced stats were more skeptical of his stardom. The Celtics had a better net rating with Brown off the court than on it in seven of his 10 seasons in Boston, including each of the last four. (All on/off data in this piece comes from Cleaning the Glass, which strips out garbage time.)

With Brown poised to play against his old team for the first time on Tuesday (8 p.m. ET, NBC/Peacock) -- and with the NBA's best current rivalry poised to begin a new chapter -- it's the perfect time for a deeper dive into what exactly the gap between Brown's reputation and advanced stats means, and how it contributed to the summer's blockbuster trade.

Given the statistics explosion in the NBA's ongoing information era, it's easy to access on/off numbers for every player in the league. Based on conversations with both public and team-side analysts, here's the best way to understand how to interpret them.

Jump to topic:
The on/off basics
Interpreting the use of on/off
How on/off contributed to Brown's trade

The on/off basics

A modern NBA front office is inundated with data about every player in the league. Points, rebounds, assists, shot quality, pick-and-roll frequency, isolation efficiency, jump shot release point, maximum footspeed, average body orientation on closeouts, and on and on and on.

The list of individual data points is nearly endless, but in theory, they all serve the same goal: A front office should identify players who can help the team outscore its opponents, rather than players who collect strong statistics in a vacuum.

Often, the players with the best individual stats also provide the most winning impact -- but not always. And those occasional mismatches can yield different and sometimes counterintuitive evaluations of players.

Consider the highly drafted duo in Orlando's frontcourt. Former No. 1 pick Paolo Banchero has more career points, rebounds and assists per game than teammate Franz Wagner. But Banchero has a negative or neutral on/off split in every season of his career -- meaning the Magic have always been better with him off the floor -- whereas Wagner has finished with a strongly positive on/off split every season.

Most of that gap comes on defense: The Magic's defensive rating has been 3.8 points higher with Banchero on the floor across his tenure, while it has been 5.4 points lower with Wagner on the court. Wagner doesn't generate many steals or blocks, but the idea behind on/off data is that he's clearly doing something right on that end, even if it doesn't show up in the box score.

Ask folks around the league about the profile of a player who's more valuable than his basic stats suggest, and they'll often talk about defensive standouts. That group includes players such as Alex Caruso and OG Anunoby, who have played major roles for the past two champions. They've followed in the footsteps of Shane Battier, famously dubbed the "No-Stats All-Star" in a 2009 piece by author Michael Lewis.

It's no surprise that those role players posted exemplary on/off numbers throughout their careers. But that reputation can take a very long time to stabilize. An analyst for an Eastern Conference team said that 10,000 possessions, or about two seasons of data for a starter, is the minimum sample size before on/off data starts becoming useful.

That caution runs counter to the incessant need for information and analysis, even when the new season's samples are unsuitably tiny in early November. For instance, it's tempting to look at Amen Thompson's early on/off data point for the 2026-27 season and think there's something wrong with his development: The Rockets have been a whopping 47.3 points per 100 possessions worse with him on the floor through nine games, by far the worst mark for any rotation player in the NBA.

But experts know that's the wrong way to interpret the data, and they haven't come close to downgrading their opinion of Thompson's value yet.

"The public should not have access to on/off this early in the season," the East analyst said. "It should be banned. I think it's a disservice to the discourse."

Transforming on/off data from 'astrology' to analysis

The problem with on/off data in its raw, preliminary form is that it's subject to all sorts of inherent biases, including that sample size issue.

"I call it astrology for NBA professionals," said an executive for a Western Conference team, "because whatever story you want to tell, you can go and [find confirmation]."

For this reason, professional analysts say they don't typically work with on/off data until they've made a host of adjustments.

To transform on/off data from astrology to analysis, they employ an advanced statistical technique called regularized adjusted plus-minus, or RAPM, which adds in various adjustments to refine the raw data. This process "does a much better job of understanding context and isolating the player's actual impact," said Taylor Snarr, a former analyst for the Jazz and creator of the advanced stat estimated plus-minus (EPM).

Depending on the specific model, RAPM controls for factors such as:

The identity of the other nine players on the court (four teammates and five opponents)

Home-court advantage

When teams are in the bonus

The opponent's 3-point and free throw percentages, which aren't affected by individual defenders

The "rubber band effect," which shows that teams play worse when they're ahead by a large margin (like a rubber band snapping back)

Whether a defense is playing in front of its own bench, which can improve performance

The most important is the first bullet point, largely because it can account for situations in which the same players "are playing together frequently and consistently," Snarr said.

Two years ago, for instance, Christian Braun finished with the third-best on/off swing among rotation players -- but Nikola Jokic finished first, and Braun shared the court with Jokic for the vast majority of his minutes. Without properly adjusting for his teammates, it would look as if Braun -- who is still a solid contributor but not in the elite echelon -- was one of the NBA's most valuable players.

For advanced player value models such as EPM, the next step is to incorporate what's called a "box score prior," which means blending box score statistics into the adjusted on/off mix to give credit to the right players. (Some models also incorporate various amounts of advanced tracking data at this point.) In that Nuggets example, Jokic's box score stats are far superior to Braun's, so the model can be more confident that Jokic is the main driver of that extraordinary on/off split.

These adjustments allow for much quicker, more accurate analysis of players because of how they isolate impact. "Instead of taking 10 years for on/off to be convincing, you can reduce it down to maybe two years with RAPM," said Kostya Medvedovsky, creator of the DARKO projection system.

This general process is now fairly routine for NBA model builders. But it can be more of an art than a science, and the trickier question is "how heavily do you prefer the [box score] statistical components versus the on/off components," the Western executive said. "That's where I think there becomes more of a legitimate debate and variance amongst the nerds."

For a player like Jaylen Brown, whose traditional stats far outshine his on/off impact, that debate plays an outsized impact in his evaluation. If a player value model weights box score stats heavily, then it will likely conclude that Brown is an elite player because of his strong scoring and all-around production.

But if a model weights box score stats less and on/off data more, then it won't be as impressed with the former Finals MVP. Brown's on/off impact on Boston's offense was typically mediocre because he had poor efficiency for a high-usage player and he posted too many turnovers to too few assists.

Public RAPM-style metrics generally consider Brown to be a very good player but not an MVP-caliber alpha. That mismatch is just one reason that his trade was so controversial.

How on/off might have contributed to the Brown trade

It's hard to deny that the proliferation of on/off-based metrics in both the public discourse and NBA front offices has transformed the league over the past couple of decades.

"I think what's changed the most is the acceptance among nontechnical people," the Western Conference executive said. "Not because they've sat down and taken in those lessons [from analytics], but just because the way everyone understands players has shifted, I think because of the success of those kinds of metrics in helping teams identify and evaluate talent."

It helps that many of the best-run franchises clearly abide by the principles of adjusted on/off data. Executives point to the Celtics and Thunder as two of the teams best known for "RAPM maxxing," which is observable from the kinds of players those teams tend to acquire. Derrick White, Isaiah Hartenstein and Caruso have always rated well by adjusted on/off metrics.

In fact, Caruso's teams have been better with him on the court in all nine of his NBA campaigns, across three separate organizations and two championships. There's a reason both DARKO and EPM rate Caruso as a top-30 player in the league on a per-possession basis, even though he has never averaged more than 10.1 points per game in a season.

Granted, not every RAPM-influenced move is going to be a slam dunk. Analysts point to the Thunder's acquisitions of Patrick Patterson in 2017 and Gordon Hayward in 2024, who were both RAPM darlings before busting in Oklahoma City. Patterson notably posted double-digit on/off swings in the positive direction in his last two seasons in Toronto, right before posting double-digit on/off swings in the negative direction in his two seasons in Oklahoma City.

But those unsuccessful acquisitions didn't stop the Thunder from continuing to trust their models, rival analysts say. And it's that mindset that led them to add Hartenstein and Caruso in the same summer. The Thunder were champions a year later, with Caruso maintaining his positive on/off streak in each season with Oklahoma City so far.

Another player with positive on/off splits in every season of his career is Paul George, across his stints in Indiana, Oklahoma City, LA and Philadelphia. Just one glance at this graph comparing the yearly on/off splits for George and Brown shows why Boston might have preferred to trade the latter for the former, even before considering financial considerations, draft picks or Brown's behind-the-scenes fit in Boston.

Many league insiders were still taken aback by the Celtics' return for an MVP candidate in his prime when they dealt Brown to Philadelphia for George -- who, despite his impressive on/off track record, is older, injury-prone and highly paid -- and picks over the summer. But it wasn't exactly a secret that Boston wanted to trade Brown, so other interested teams had the opportunity to beat the 76ers' offer if they thought signing the Finals MVP was worth a higher cost.

That no better offer came demonstrates just how deeply the lessons of adjusted on/off metrics have permeated the thinking of most front offices.

"In some ways, the Jaylen Brown case is the purest proof of this," Medvedovsky said. "It seemed like: You're trading a Finals MVP, so you're going to get a ton of value for him. But then we saw you just didn't."

This wasn't a case of one ahead-of-his-time, outside-the-box general manager making a shrewd move; rather, it was a case of 28 other general managers not making an all-in offer for a player with underwhelming adjusted on/off metrics.

Now in Philadelphia, however, Brown is attempting to prove that there should have been a massive market for his services: He's averaging 21.6 points per game while often serving as the third or fourth option on the East-leading, 8-2 76ers.

And while it's far too soon to come to any meaningful conclusions about Philadelphia's performance splits with and without him -- Brown has played just 678 possessions as a Sixer, per Cleaning the Glass, or less than 10% of the way to 10,000 -- one early-season data point is mighty compelling.

With Brown on the floor, the 76ers have outscored their opponents by 4.3 points per 100 possessions; without him, they're at minus-1.9. That 6.2-point swing, if sustained, would represent the second-best single-season on/off of Brown's career.

"With [Daryl] Morey gone," Medvedovsky added, "there's no GM who I would identify as, this is the stats GM. But in some ways, it's like a total victory for the on/off data that even with no visibly stats GM, even the non-stats people were on/off-pilled enough that there just wasn't that kind of massive market for Jaylen Brown."