In NBA 2K18, each active player in the NBA, along with a healthy number of legends, is individually rated on a 99-point scale. When the game launched on September 19, LeBron James was a 97. James Harden was a 95. DeMarcus Cousins was a 92. Lonzo Ball was an 80.
And every year, these ratings are held up to scrutiny—sometimes by the fans and critics, but most often by the players themselves. Emotions run high. John Wall, who was rated a 90, called out 2K marketing director Ronnie Singh on Twitter. Rudy Gay, who was rated an 83, called the rankings a "crock of BS."
But the process of rating these players is not some backstage popularity contest, where the developers assign final ratings based on their gut instinct. It is a scientific process intended to eliminate as much bias as possible.
By digging through the menus, one can see that a player's overall rating breaks down into several smaller ratings. James Harden, for example, has a passing accuracy of 78; this is a drop from 2K17, because last year, he led the NBA in turnovers per game. His durability is 95. His free throw is an 85. His ability to draw fouls is a 98. The developers consider approximately 50 such attributes to get a complete picture of each player. And this can get hairy, especially since many attributes complement one another and might inadvertently buff or nerf a player.
"We have to balance the statistics," said 2K Games Producer Michael Stauffer in an interview with Complex. "For example, a guy like Kyrie Irving has great lateral movement. We don't want him to feel slow, but at the same time, we don't want him to be overpowering on defense."
Stauffer and his team of developers have one of the most important jobs at 2K Games—to manage the raw player data under the game's 4K graphics hood. It is a process that is both time-consuming and exhaustive. Now there are so many different types of player builds; you get a primary and a secondary archetype, which determine the upgrade caps for your skills. The formula must account more for what type of player you want to be, rather than than just the player's position. A playmaking (secondary) defender (primary) point guard is going to be evaluated differently than a defensive (secondary) playmaker (primary) point guard.
A player's steals per game or his mid-range shooting percentage will correlate to his in-game rating in those respective categories. This commitment, of tying a player's in-game rating to hard, real-life data, is omnipresent. Durability is determined by a player's injury history, right down to the individual limbs and joints. Pass consistency is based on assists. Passing accuracy is determined by turnovers.
Beginning in 2016, the NBA began tracking real-life "hustle stats," tallying the number of screen assists, contested jump shots, and charges taken per game, among other stats. And now, ratings for more subjective categories like IQ or Strength, which always require a bit of guesswork, have additional, concrete support.
Some ratings, however, do require a bit of human involvement. Take speed, for example. The developers have stats from the NBA Draft Combine, which clocks a three-quarter-court sprint. But there is no way to compare this to a player's top speed while dribbling the ball. So Stauffer and his staff will watch hours of archival footage to procure an accurate rating. And archival footage is especially important to rate NBA legends like the late Wilt Chamberlain, who played before statistics were so intimately tracked.
Proper context for the numbers is important as well.
"If a player has a 37 percentage for his 3-point shot, we're interested in 'how,''' Stauffer said. "Not every 3-point shot is the same. There's always a variable. For example, if Steph Curry has to run across the whole court through multiple screens to find an open 3—or has to shoot from far back to get open—does his percentage tell the whole story?"
And some statistics only matter if the AI is controlling the player. Russell Westbrook has a Pick and Roll Defense IQ of 85. which only determines the computer-controlled Westbrook's IQ; if you're playing as Westbrook and you generally have trouble dodging and predicting screens, the IQ rating won't make things easier.
But even after all these numbers are litigated and finalized by the developers, the final rating is still subject to change. Numbers alone cannot tell the entire story.
"It's a lot about feel," Stauffer said. "Does it 'feel right' when you're playing the game? Is this player a bit faster than what we see on TV? It's a lot of trial and error and a lot of discussions in the office to make sure this is as accurate as possible."
"We monitored trends during the off-season, specifically for the 3-point shot," Stauffer said. "The league has become such a 3-point league, and you'll read articles about guys that are trying to add that shot to their game. So we'll give a small bump to a player like that."
Cousins, for example, who began developing his 3-point shot two years ago, now has a 3-point rating of 80.
If a player isn't happy with his rating, 2K Games does update the ratings regularly to reflect changes in performance. It doesn't happen daily—a single game's fluke is not going to change a player's entire rating, for better or for worse. But it will happen every 2-4 weeks, so that players can improve their rating with a consistent streak of strong turnouts.
"We don't want the players to be upset," Stauffer said. "Obviously, they take a lot of pride in their work. And so if a player is unhappy, we'll double check our work and make sure that everything checks out."
"Players come by the office, and it's fun to look at the ratings with them and get their feedback," Stauffer continued. "At the end of the day, they're the experts. They're the ones who are actually playing."