Introducing xPTS-RAPM

Ayomide Awoyemi · Luma Basketball Research · 16 September 2026 · Original thread

Good day to you reader. I hope you are having a good day as you stumble upon this paper. My name is Ayo and, as much as I love basketball, I am Nigerian, thus soccer, or rather football is in my blood. If you tuned into any World Cup games this year, one of the things you'll see repeatedly on game summaries and even in missed games is expected goals per team.

I want to take some time and write a quick blog explaining a new stat that I tinkered with and created that borrows from the concept one of the foundational analytics in football, that being xG.

What Is xG?

image sourced from: https://statsultra.com/xg-explained-what-are-expected-goals/

xG is football's statistical measure of a shot as it relates to its goal. Given the skill of an average player, how good of a goal opportunity was a shot? It strips away the luck and skill differentiation and focuses on the process of the shot in isolation.

Every shot is given a quality grade between 0 and 1, with 1 representing a guaranteed goal. A shot graded with an xG of 0.10, means over 100 times it would be statistically expected to be a goal 10 times (1/10). For added reference, a penalty kick typically has an xG of about 0.79.

Watching this World Cup reminded me of a conversation I had with my best friend from when we were watching the Man U game about how cool it would be to bring the concept of XG to basketball. I spent a week-ish tinkering with models and created xPTS, my iteration of basketball's version of xG.

What Is xPTS?

xPTS are the expected point valuation of each field goal and free throw attempt compared to league average.

It grades each shot attempt on a scale from 0 to 3, based on how many points that shot would be expected to produce if taken by a league-average NBA player, stripping out luck and individual conversion skill.

xPTS(shot) = league PPP of the shot's bin (zone × distance × shot family × possession context)
xPTS(free throw) = league avg FT% × 1
Total xPTS = Σ bin value of every field goal attempt + league FT% × free throw attempts

Each field goal attempt is placed into a bin based on four characteristics:

• Court zone: 14 zones
• Shot distance: two-foot increments
• Shot family: drive, cut, catch-and-shoot, self-created, alley-oop, or putback
• Possession context: halfcourt, transition, second-chance, or off-turnover
• Defender Distance (Sports VU leak, more on this below)

This codified approximately 707 bins per season across roughly 219,000 field goal attempts.

Here is a sampling of bins across the value spectrum.

xPTS-RAPM

Instead of seeing how a player's on-court presence adds to or subtracts from a team's net rating over a huge sample of possessions, you see how a player's on-court presence influences a team's expected point value.

Are they creating numerically optimal shots? Are they suppressing numerically optimal shots? How are they warping the geometry of the court?

Like a normal RAPM, you can then decompose this further into offensive xPTS-RAPM (oxRAPM) and defensive xPTS-RAPM (dxRAPM): how a player changes the geometry of the court to bend his team toward more favorable shot conditions, and how his presence helps suppress favorable shot conditions on the other side.

best offenders and defenders below.

As you may notice, Donovan Clingan and the Thompson twins are amongst the most surprising and notable leaders, especially since they are not good three-point shooters, nor are they playmakers. I will have to do more decomposition to see, numerically, why they rank so well.

These leaders are actually a great case study of what this RAPM measures. Clingan is a chance-quality machine in three distinct ways.

His rebounding: 22.2 offensive rebounds per 100 shots is a good 30% more than league average. Many of those rebounds convert a dead possession into a putback or a kick-out, which sit amongst the most highly priced bins, around 1.3 to 1.5 expected points.

And his rebounding lets Portland manufacture its diet: with him on the court, Portland takes 49 % of its shots from three and 32% at the rim, while opponents get only 26% of their shots at the rim. Portland's offense with Clingan on is extremely Morey-pilled: rim attempt or three-point attempt.

For the Thompson twins, Amen is actually the best offensive geometry-bender in the league (bar Giannis). He is the offensive xPTS engine: extreme rim pressure and second chances. Remember, with expected points there is no tax, no punishment for his lack of conversion, or his team's, while he is on the court.

Ausar, meanwhile (5th in xPTS RAPM), is elite defensively. His presence lets his team take 35.8% of its shots at the rim while dropping opponents' rim share to about 24.3%, and he forces about 19 turnovers per 100, zeroing out those possessions entirely.

On the inverse, you have the players who perform extremely poorly in net xPTS-RAPM.

The worst being Bub Carrington, and the interesting point about him is that there is no single horrific channel.

He has, up to this point, been poor among so many.

Washington's offense becomes extremely midrange-heavy when he is on the court. His personal game is heavily pull-up-weighted with no rim pressure, which drags Washington into the cheapest quantified shot values.

He provides no transition boost, below-average offensive rebounding, and above-average turnovers, essentially never generating or forcing his team toward more optimal chances while leaking possessions along the way.

33.9% of opponents' shots come at the rim with him on, against a league average of about 28.7%, while opponents take only 25.8% from midrange, the exact inverse of what a good defense produces.

He does not create turnovers. He gives up straight drives to the rim, and opponents' free-throw share runs above average.

Some honest context, which I will return to in the caveats: he plays for a 17-win team that has been bad as an entire unit, he is a young guard given developmental freedom and reps, and the collinearity weakness of any ridge regression does exist.

The beauty of xPTS-RAPM is that it is subtly a decomposition channel of RAPM itself: one half of a shot-quality decomposition.

xPTS-RAPM is the half that measures how many chances a player creates or suppresses through the geometry of the floor.

Its sister, Conversion RAPM, measures how much his presence helps convert, or deny, shots above expected.

The summation of the two gets you Net/Pure RAPM, the standard RAPM we all know and love, the one that measures how team net rating moves with a player on or off the court.

Conversion RAPM

xPTS-RAPM: a quantification of how a player influences his team's shot diet.

Conversion RAPM: a quantification of how a player's on-court influence alters the points created/suppressed above/below expected.

Net RAPM

After running numeric validation, I found that I had somehow built, from first principles, a full reconstruction of pure RAPM.

Using independently derived expected-points measure for shot values, and the conversion above or below those shot values, adding the two regularized adjusted plus-minuses together gives you a number that is 99.99999% equivalent to a pure, no-prior RAPM.

This is the part I am extremely excited for, standard pure RAPM asks what happens to the score: how does net rating change with this player on and off the court?

xPTS and Conversion take that question and rip it into two further questions.

How much does this player change his team's quantified shot diet?

And how much does his presence swing his team's conversion above or below that shot diet?

Essentially, separating regularized adjusted plus-minus into shot-quality impact and conversion impact.

The identity requires the same RAPM specification in all three solves: the same stint rows, lineup matrix, stint weights, centering, and ridge penalty. Another researcher can use different stints or a different xPTS model and still recover the same identity inside that specification, as long as conversion is defined as actual points minus expected points and all three solves use the same settings.

Creation, Conversion, and Which One Repeats

In football, xG is a better tool for prediction than actual goals, as conversion in football is lossy and does not repeat. Conversely, in basketball, conversion is one of the defining skills, while “chance generation” is compressed. (Once again especially without defender distance)

Conversion It is the critical context that separates an average team/player from the elite.

NBA over/under-performance of xPTS repeats year after year: r = 0.55 (120 team-season pairs) And .58 at the player level.

EPL (English Premier League ) finishing over expected goals repeats at r = 0.12. at the player level.

(Source:Pipping-Gamón, J., Feng, T., and Sabin, P. (2026))

So for basketball the creation of xPTS, or any form of expected points, cannot really be used to predict a team's future performance. It is added context for how a team gets to its end result, and it is, more than anything, a confirmation of how critical shot conversion is. Tough shot-makers, and shot-makers in general, carry an incredible amount of value.

Kevin Durant and the value of tough Shot Making

Kevin Durant and net xPTS-RAPM is a very interesting story, especially when you pair it with his conversion.

His shot-quality impact is among the single worst in the entire league. He provides no rim pressure, there is no transition game, the offense becomes a half-court jousting match every possession. His influence turns his team into a midrange engine, the quantified worst shot in basketball, at roughly 7 percentage points more than league average.

He still rates well in all RAPMs, because he is, even at his age, among the most elite converters in the sport.

He makes the poorly graded 8-to-12-foot pull-ups at about 1.25 points per shot versus a league average of 0.92, functionally making what is graded as the worst shot in basketball worth more than a three from a league-average shooter.

He ranks in the 1st percentile of the league in chance creation and the 100th percentile in conversion.

Extreme dimorphism.

And that, as I will keep touching on, is the beauty of xPTS and xPTS-RAPM. You can now decompose what RAPM even stands for in a new way: how many chances a player creates or suppresses through the geometry of the floor, versus how much he converts himself, or erases.

Caveats

• The 2015–16 work is unfinished. The SportVU leak gave me roughly 25% of that season's defender-distance data, 295 of 1,230 games. Everything built on it is a projected sample, not a season.
• The main formulation has no defender distance. The 707-bin engine captures about 77% of the value found in defender distance. I lose roughly 23% of real shot-quality information, at the lowest. My xPTS is unfinished until I get real defender-distance data. Treat this as a working proof of concept.
• The bins are trained per season. Each season's 707 bins use that season's shots only, shrunk toward the parent zone, recomputed every year. Free throws price at the season's league rate, not the player's own.
• This is shots only, not full EPA. There is no passing value in the quantification. xPTS is expected points per shot attempt.
• The collinearity problem persists, even in xPTS-RAPM. Teammates who always share the floor are hard to separate. Some players carry credit that belongs to their lineups. 2,000 weighted possessions is a floor, not a cure.
• Descriptive, not predictive. Chance creation does not out-predict actual results in basketball. xPTS and its RAPMs are context for how value is created, not a forecast.

Fin

Every leaderboard in full, the interactive tables, and the complete methodology live at court-share.com/luma/papers/introducing-xpts

Is this thing on? #NeverBooz #Dybansta1

Full 2025-26 leaderboard