How the 54 skill grades on the Player Breakdowns wheel and the player cards are built, how the similar-players list works, and the conventions the rest of the site assumes you know.
Part 1 · Skill Grades
Every player-season since 2016-17 gets 54 skill grades, collected into 18 groups: 12 on offense, 6 on defense. Each grade is a number on one scale: 0 is league average and +1 is one standard deviation better, measured against that season’s qualified players (1,000+ possessions). There is no cap. A +5 means the player is genuinely five standard deviations clear of the field, not that a scale ran out.
The player cards show the same grades as percentiles among those qualified players, because a 0–100 bar is quicker to read at a glance. The breakdown wheel shows the numbers themselves, so you can see how far ahead a player really is.
The data comes from four places: the play-by-play (every shot by its label, every assist, steal, block, foul and turnover), NBA tracking (drives, catch-and-shoot, closest defender, hustle stats and defensive matchups), Synergy play types, and two of the site’s own on-court models for things no box score can see.
Most skills are a rate with attempts behind it: layups made per layup, points per isolation, turnovers per pass. All of them go through the same four steps:
Shrinking keeps a small sample from leading a leaderboard. For shots, k is 25 attempts and it fades out completely by 150, so a real season is graded on what actually happened. Noisier things get pulled harder:
The expectation is what makes a grade fair, and it changes with the skill:
Volume is measured per 75 possessions against the median player:
Because the score is efficiency times volume, three things follow:
Volume counts for more where carrying the load is the skill: up to 3× for isolation, pick-and-roll and post-ups, and fully for deep threes, lobs and rim contests.
| Kind of skill | How it is graded |
|---|---|
| Counting skills | Things with nothing to divide by: assists, steals, blocks, deflections, charges, screen assists. The per-75 rate is shrunk toward the league rate (1,000 possessions of average, fading out by 3,000), standardised, then pulled toward 0 for players who have not played much. |
| Two-part skills | Contact Finishing, both rebounding skills, Boxing Out, Closeouts, Passing Lanes and Isolation Defense blend two measures. Each part is standardised first, so a weight such as 60/40 means what it says. |
| Doing it is the skill | Dunking and Lob Threat are judged against a layup, not against other dunks. Nearly every dunk goes in, so the grade becomes how often he gets there. Not dunking at all is a low grade. |
| Matchup skills | Isolation and Post Defense, Positional Versatility and Guarding Stars come from tracked defensive matchups, which start in 2017-18, so 2016-17 has none of them. |
| On-court models | Rim Deterrence and Transition Defense come from the site’s own RAPM-style regressions, which already shrink their estimates. They are only standardised. |
Under the recipe, a player with a handful of attempts lands close to 0: there is not enough evidence to call him good or bad. That is right for a skill he barely uses. It is wrong when he does not use it at all, so the site leaves those grades out:
A blank skill drops off the wheel and out of its group’s average.
A group grade is the average of its skills, re-standardised among qualified players so it reads on the same scale as a single skill. Two groups are built differently, because their skills are slices of the same shots:
There is no overall Offense or Defense grade. The centre of the wheel is coloured by the average of its groups so it reads at a glance, but no number is printed. One number would hide exactly what the wheel is there to show.
Unless a row says otherwise, a rate skill uses the recipe above and a per-75 skill uses the counting rule.
Offense · 12 groups, 37 skills
| Group | Skill | What it measures |
|---|---|---|
| Finishing | Rim Finishing | Half-court layups he created himself (driving, reverse, finger rolls). Dunks, cuts, transition and putbacks are other skills. FG% vs his position. |
| Contact Finishing | On every rim finish: how often it is an and-one, minus how often it gets blocked. Both vs his position. | |
| Above the Rim | Dunking | Dunks, judged against a layup rather than other dunks, so the grade is mostly how often he gets to one. No dunks is a low grade, not an average one. |
| Lob Threat | Alley-oop finishes, built the same way as dunking. | |
| Touch Shots | Floaters | Every floating shot, vs his position. |
| Hooks | Every hook, including turnaround and driving hooks, vs his position. | |
| Short Jumpers | Two-point jumpers from 14 feet in, not counting turnarounds and fadeaways. | |
| MidrangeAttempt-weighted | Pull-up Midrange | Pull-up twos, against what that shot should go in at given how closely it was guarded. |
| Catch-and-Shoot Midrange | Catch-and-shoot twos, defender-adjusted the same way. | |
| Turnarounds & Fadeaways | Every turnaround and fadeaway two. Shown on the wheel but left out of the group: these shots are already inside his pull-ups. | |
| Three-PointSum of the four | Catch-and-Shoot 3s | No dribble, ball held two seconds or less. Defender-adjusted. |
| Quick 3s | One or two dribbles: side-steps, one-bounce pull-ups, attacking a closeout. Defender-adjusted. | |
| Off-Dribble 3s | Three or more dribbles: isos, step-backs, deep pick-and-roll pull-ups. Defender-adjusted. | |
| Deep 3s | Threes from 27 feet and out, heaves removed, each judged against league 3P% from its own distance. Volume counts in full: taking lots of them at a decent rate is the skill. | |
| Free Throws | Free Throw Shooting | FT% vs the league, weighted by attempts. |
| Foul Drawing | Share of his scoring trips that end at the line, vs his position. | |
| Self-Creation | Isolation | Points per isolation possession vs the league. |
| Pick-and-Roll Ball Handler | Points per pick-and-roll possession that ends with him (shot, free throws or turnover). | |
| Post Scoring | Points per post-up possession. | |
| Driving | Rim Pressure | Points per drive vs the league, times how often he drives. |
| Drive & Kick | How often a pass out of a drive becomes an assist, times how many he makes. | |
| Off-Ball Scoring | Cutting | Points per cut. |
| Movement Scoring | Points per possession off screens and handoffs, threes and midrange alike. | |
| Transition Scoring | Points per transition possession. | |
| Playmaking | Creating for Others | Potential assists and free-throw assists per 75: every pass that led to a shot, so his teammates’ shooting does not decide it. |
| Creation Quality | How much better his passes make a teammate’s shot, in effective FG points, times how many shots he creates. | |
| Rim Feeding | Assists on layups, dunks, tips and lobs per 75. | |
| Shooter Feeding | Assists on made threes per 75. | |
| Connective Passing | Secondary assists (two thirds) and passes made (one third) per 75. | |
| Transition Passing | Assists within eight seconds of a defensive rebound or live-ball turnover. | |
| Ball Security | Passing Security | Bad-pass turnovers per pass, against the risk his mix of creative and routine passes should carry. Higher is safer. |
| Handle Security | Ball-handling turnovers per dribble, vs a half-position, half-league rate. | |
| Avoiding Offensive Fouls | Offensive fouls per touch, the same way. Moving screens are in here: the play-by-play does not separate them. | |
| Screening & Glass | Screen Setting | Points scored off his screens per 75. |
| Rolling | Points per possession as the screener, rolling or popping. | |
| Offensive Rebounding | Share of his team’s misses he grabs (60%) and how often he wins a contested board when he is near one (40%). | |
| Putback Scoring | Points per possession after an offensive rebound. |
Defense · 6 groups, 17 skills
| Group | Skill | What it measures |
|---|---|---|
| Perimeter | Isolation Defense | Points his guard and wing matchups score on him, against what the same players score on everyone else; isolation possessions allowed add up to a quarter of the weight. |
| Closeouts | Three-point contests per 75 (65%) and spot-up points allowed (35%). | |
| On-Ball Stripping | Steals off lost-ball turnovers per 75: picking the dribbler. | |
| Perimeter Blocking | Blocks on threes and jumpers from 15 feet out, per 75. | |
| Interior | Shot Blocking | Blocks inside 15 feet per 75. |
| Rim Contesting | FG% allowed within 6 feet, against what those same shooters normally hit there, times how many rim shots he contests. | |
| Rim Deterrence | How many fewer of the opponent’s half-court shots come at the rim with him on the floor, adjusted for everyone else on the court. | |
| Rebounding | Securing Boards | Share of opponent misses he grabs (60%) and contested boards won per chance (40%). |
| Boxing Out | Defensive box-outs per 75 (70%) and how often his team gets the board after one (30%). | |
| Disruption | Passing Lanes | Steals off bad passes and deflections per 75, half each. |
| Charge Taking | Charges drawn per 75. | |
| Loose Ball Recovery | Loose balls recovered per 75. | |
| Assignments | Post Defense | Points his big-man matchups score on him, against what the same bigs score on everyone else. |
| Positional Versatility | How evenly his matchups spread across guards, wings, forwards and bigs. A measure of role, not of how well he guards them. | |
| Guarding Stars | The average quality of the players he guards, by usage and scoring. Also role, not results. | |
| Discipline | Fouling Discipline | Defensive fouls per contest, vs a half-position, half-league rate. Fewer is better. |
| Transition Defense | How many fewer of the opponent’s chances after a defensive rebound become a transition shot or free throws with him on the floor. |
The grades are built up from individual skills, so a fair test is whether they agree with a measure built from the top down. Averaging the twelve offensive groups over 2023-24 to 2025-26 and comparing that with three-year offensive RAPM over the same seasons, for 271 players with 3,000+ possessions, gives a correlation of 0.69. The groups that track impact best are Self-Creation, Finishing and Free Throws. Above the Rim tracks it slightly negatively.
Where the two disagree, it is mostly in two places:
The card shows six groups a side so offense and defense line up row for row:
Under each side is an Impact row from the player’s RAPM, ranked against the same players. The full wheel is one click away from every card.
The ramp goes from red through a neutral grey to green rather than straight across, because a direct blend passes through olive and would make every average player look muddy. Average looks colourless, and only clear strengths and weaknesses carry colour. It is at full strength by ±2, where the number keeps the rest of the story.
Part 2 · Similar Players
The similar-players list on a player card answers “who plays like this?”, not “who is as good as this?”. A bench player and a star with the same stylistic fingerprint are a strong match, and that is the intended behaviour.
Each player becomes one vector of nineteen numbers:
Every field is z-scored across the qualified league pool for that season before the comparison. Without that step the high-variance zones would swamp the rarer play types entirely, and the list would quietly become “who shoots from the same places” with the play-type half contributing nothing.
The five highest scores are shown. Note that this is a different tool from the role templates: roles match a player against a fixed, hand-authored archetype, while this matches him against every other real player in the league that season.
Part 3 · Conventions
Relative true shooting appears three times, and the only difference is the pool it is measured against. All three are minutes-weighted.
| Stat | Compared against |
|---|---|
| rTS | League-average TS% that season. |
| rTS Pos | Players at the same position that season. |
| rTS Role | Players sharing his primary scoring role that season. |
A rank is always within the qualified pool for that season, and the total is shown beside it so the denominator is never a guess. For stats where lower is better — opponent FG% allowed, turnovers — rank 1 is the lowest value, not the highest.
A corner three is a shorter, far more frequently assisted, genuinely different shot from an above-the-break one, so that split earns its place. Left versus right for the same player is close to pure noise, so those are pooled — which roughly doubles the sample behind every zone estimate.
A multi-year window is named for the season it ends on, so a two-year window labelled 2025-26 is built from 2024-25 and 2025-26. Anywhere a rating and a single season sit side by side, the rating partly reflects the year before.
Skill grades, role assignments, role-relative baselines and similarity are all computed inside a single season’s player pool. Nothing is fitted on one year and applied to another, which is why an in-progress season is internally consistent from its first weeks — it is being compared to itself, not to a finished year.