Playbooks & Play Type:
- Major discoveries :
- Some playbooks are bugged for years, such as the Mavs one, and no matter the settings, the same freelance play will be called infinitly. We have to replace them by using user or custom playbooks slots.
- Some plays are pre-scripted and remove the decision making of AI, they have to be identified and removed.
- Some plays are slow, inefficient, or include very bad spacing or weird motions, they basically kill the flow of the game and have to be removed as well.
- Calibration & Update :
- We identified and backtested more than 800 unique plays, efficient, with modern motion and spacing, on a AI point of view.
- We designed all our playbooks based on those 800 filtered plays, with some constraints such as who is the final passer for handoff, does the PF need to be waiting in the corner like A.Gordon for the Nuggets, …
- A special treatment has been done on isolation plays to identify plays replicating the mismatch hunting, with several potential outcomes such as drives, go to shot, kick out.
- Play types are designed by on nba.com/stats data.
Attributes :
- Major discoveries :
- The YouTuber 2K Tutes has tested and identified the impact of each attribute and their value range (
https://youtube.com/@nba2ktutes?si=912CZwxH3xiXsTvP).
- The on ball defence is quite complex, as several attributes have significant impact :
- Perimeter defence and strength has the same impact in terms of getting caught off animations.
- Defensive consistency is the strongest defensive attribute as it defines the probability to react perfectly to any situation (rotation, contest, block, ..). Lower values here will allow more fouls as well.
- Vertical is an issue, as it does not consider the height of players, so a big like Gobert with high vertical can basically jump for the paint to the 3pt line and contest heavily any shot. It impacts also rebounds and therefore rebounds distribution over a team.
- We have identified the impact of each 3pt attribute value and the FG% it gives over sim games in MyNBA.
- Passing IQ is vital for ball movement, and needs to have a high range of value to improve AI decision making, pace and motion. On the other hand, vision will help to seperate the great playmakers versus bad ones.
- The value range of Pass perception needs to be really high to allow more turnovers and pass deflections/steals.
- Stamina needs to be max out to have realistic FG% over MyNBA.
- Calibration & Update :
- Defensive consistency : We have to cap this attribute and lower the range of potential values so that even the best defenders in the league can react imperfectly sometimes. It allows also more help, more mistakes during rotations, and a much better motion overall.
- Vertical : our algorithm defines appropriate ranges of value by height.
- Our 3PT rating attribute is based on career 3PT FG% to avoid the huge impact of volatility year on year, and based also on the 3PT FG% given by each value over sim games and gameplay.
- Our defensives attributes have been calibrated in such ways that we are able to create several archetype of defenders, among them we have : good 1vs1 defenders in the post, great perimeter defenders based on lateral movement or strength, good stealers/interceptors, good help IQ defenders, complete defenders. For that we calibrate the following :
- Perimeter defence is based on match up data versus the top 15-20 isolation players in the league (regular season and playoffs), where our model considers : how often the defender is guarding the top isolation player/min, forced turnover/min, block/min, defensive FG%/, fouls conceded/min, how many good defenders a team possesses (because data can be weak in the case of Boston where they have several good perimeter defenders).
- Interior defence is based on defensive FG% data in the paint for the past 2 years.
- Help defence is based on a mathematical model using the following normalised data : deflections/min, loose balls recovered/min, charges drawn/min, contested 2pt/min, contested 3pt/min.
Tendencies :
- Major discoveries :
- There are 3 types of tendencies (gameplay only, sim games only, both gameplay and sim games). When backtesting, we discovered that tendencies impacting both sim/gameplay are acting as multiplier for gameplay only tendencies. It means that their values will be used to define the possibility to trigger gameplay tendency. As an example, if you have shot 3 tendency = 20, and step back 3 = 100, the AI will barely uses the step back 3 animation. On the other hand, with a shot 3 tendency = 100, and step back or spot up 3 at 20, you will see quite some 3s and step back. Also, gameplay tendencies like spot up 3 must not be above 50, otherwise it creates really weird animations and behaviour from the AI.
- Shot tendency also acts as multiplier for gameplay only shot tendencies.
- The higher the multipliers tendencies, the higher chance of pump fake from AI for 3PT shot or mid range shot you will see (check my videos on YouTube to see the AI doing several pump fake at the 3pt line or at mid range spot or in the post).
- Contested 3 tendency defines how often the AI will use the new go to shot animation. This is a determinant tendency if we want to see players throwing 3 points shot during size up in isolation.
- Drive pull 3 tendency defines handoff 3pt.
- Offscreen 3 tendency defines 3pt shot taken after movements and screens (for Klay Thompson guys).
- Play discipline defines the possibility of AI to “break” the motion of a play, basically, with low values the AI will have more willingness to think and react to defensive rotations, accordingly to his set of attributes/tendencies. Whereas, with high value, the AI will follow perfectly the motion of the play, and we remove the decision making of AI. That’s why with very high play discipline you can see sometimes the AI in “brain dead mode”, just standing and doing nothing and throwing a bad end of clock shot. But with too low play discipline, then the flow is anarchic, as the AI is playing like street basketball, not respecting the motion of playbooks. Also, the more AI is willing to think and react, the more diverse are the outcomes of plays, it leads to more potential turnovers and better fast breaks. Therefore, play discipline needs to be calibrated accordingly (min 20, max 70) and the difference between players must not be high so that there is a good repartition of ticket shots, and we’ll see more good/bad reactions and decision making on offence when a play is running.
- Attack strong on drive must be low for the “crafty” playmakers such as Doncic, to allow him to take his time when driving, to either throw a mid range floater, send a lob, kick out or do whatever necessary to punish the defence. We define it in a range of min 20, max 80.
- Low Play discipline is the major parameter to allow AI to throw deep 3s.
- Shot close tendency represents the mid range floater animation.
- Dribble tendencies must be at 0, otherwise it kills the motion of AI during drives and fast break. AI will still perform dribble moves if necessary.
- Flashy pass needs to be at 0, those animations are terrible and old.
- Post moves tendencies must be capped at 35 so post players take their time to “work” at the post before taking a decision.
- Block and contest tendencies must not be too high for centers (typically not at 100 for every big), otherwise they contest stupidly every shot and they allow free offensive rebounds when they drop during pick and roll, instead of allowing a free mid range and defending the rebound.
- Block, contest, steal and pass interceptions tendencies must be in a range of minimum 50, maximum 100, average 75. So sometimes AI is smart to give free layup and free mid range instead of allowing and ones and free offensive rebounds, as they do IRL.
- Calibration & Update :
- All tendencies have been defined based on hours of backtesting, real NBA data, and well calibrated ranges of values to avoid any weird behaviour and allow smart AI decision making and versatility.
- Special example of how we handle the multipliers tendencies :
For instance in real-life, Durant takes 15% of his shoot from under, 23% from close, 34% from mid, 28% from 3. Those are the figures we want for the tendencies shot under, close, mid, 3pt which impact both gameplay and sim games shot distribution. But they are actually too low in absolute value, and as they act as multiplier, the AI won’t be much aggressive using specific gameplay tendencies.
Therefore, we scale and round those shot repartition tendencies with a max value of 100.
Durant will now have shot under 45, close 65, mid 100, 3pt 85. Durant will be much more aggressive in using each gameplay moves (stepback, pump fake,..) while keeping the same shot repartition as IRL.
Badges :
- Major discoveries :
- Basically, badges are killing the gameplay as they completely disrupt the balance in FG% on either offense and defence. In an ideal world, I would like to remove all badges. But there are elements that we cannot fix by tweaking only attributes, tendencies, and sliders. For instance, AI isolation offense provides low PPP.
- Indeed, the YouTuber 2K Tutes has shown that ankle breaker and physical handles animations cannot occur without those badges.
- With too many badges, it is impossible for stars to have bad days. And it makes it difficult for bad teams to do some upsets.
- Calibration & Update :
- We remove all badges, except ankle breaker, physical handle that we give to every player with silver level, so everyone can trigger the cool isolation animations if their attributes are high enough.
- We give limitless range to everyone for the same purpose, so that we can see more deep 3s and not only 3s just after the 3pt line. On the other hand, we lower the success rate of 3PT within MyNBA as limitless range increases a lot the 3PT FG% during sim games.
Sliders :
- Major discoveries :
- My mathematical models calculating attributes, tendencies consider a key element : keeping all sliders as closed as possible to 50. This is essential as tweaking sliders will disrupt the gameplay, and all players will behave in a similar way.* For some elements such as rebound, it is unfortunately impossible to reconcile sliders, attributes, tendencies between in-game and sim games.*So for those specific points, we will have to deviate slightly from 50 to balance the gameplay and keep accurate results in sim games throughout MyNBA.
Coaches & POE :
- Major discoveries :
- POE impacts both sim games and gameplay but you must already all know it.
- High help defence slider is not necessarly a good thing as AI will rotate too much and leaving a lot of corner 3s. A good balance has been found after hours of backtesting so that AI can help in the paint and also defend the corner 3s if needed.
- 2K AI is bad to set up a clear and efficient defensive scheme in-game. There is a need of manual adjustment from the User for both teams in order to have the most realistic gameplay possible. After watching a ton of NBA videos, we identified 2 majors defensive settings which are very common in NBA and we can replicate in NBA 2K : “Shrink Floor”, “Shrink Floor” with no Help (You have to manually change Help to “No” for Drive and PnR). “Shrink Floor” allows the defence to drop or switch during PnR. Also, off-ball defenders let more space to their opponent to help in the paint if needed, but they can still contest 3PT especially corner 3s. We have a slide that shows the defensive settings you have to set up for each team.*Also you have to manually change matchup to force AI to assign their best defender on your best offensive players. You can do those changes in “Gameplay”, “Defensive Settings”.
MyNBA :
- Major discoveries :
- Due to the removal of badges and lower defensive ratings, the OVR (overall rating) of players are rather low in our roster. Which is very great point for the trading system. Because AI is more inclined to trade players below 90 OVR.. We came up with a very great trading system thanks to player fame = 10 and hours of backtesting. Indeed it creates an exponential curve speaking about player value, where all-stars and stars have much more value than a random player. It makes it more realistic for User vs CPU trades and also CPU vs CPU trades. You won’t be able to fool the CPU too often now. And thanks to lower OVR and our current trade sliders, CPU is now able to identify bad contracts values. And selling teams are chasing expiring contracts !
- MyNBA sliders have been updated to provide the most accurate sim-games possible, leading to accurate players stats over a season. Also, thanks to the chemistry difficulty and chemistry impact = 100, we are able to replicate some pattern where teams can be stuck in a losing streak such as Pistons and Spurs this year. And on the other hand, it helps teams with great chemistry such as Nuggets or Wolves to stay on top of the league for a while.