PK people from the early days, before we went public, might see this piece and remember reading it back in April 2024.
But not you. You might be a dev on our indie projects, and this is new, shiny!
It'll all make sense. Read on if you're interested in game design and psychology.
Yes, we've officially kicked off indie! 3 tiny projects for 2025. Excited to share more about each one and their interplay. Let us cook.
Building the Machine
You've heard AC and I talk about "building the machine" a few times by now. That "building the machine" and watch it go is a core component of roguelike experiences.
What do we mean?
I’ve played quite a bit of Balatro (Square has too). It's great, go play it.
It's a one-dev indie project up for Game Of The Year. It toppled Minecraft's reign as top paid mobile app too.
While you play, try and analyze what’s going on in your mind. Balatro is a special playground to do so since it strips the abstractions games add on top of the mechanics.
Balatro’s mechanics are concrete: add and multiply semi-random numbers to hit a threshold.
That’s what you do in most Action RPGs, or any game where you deal a certain amount of damage over a certain amount of time to defeat an enemy. Weapon stats, armor, buffs etc. obfuscate what the core systems are about.
PENS view
In PENS/SDT (Player Experience of Needs Satisfaction)/(Self Determination Theory) Balatro hits the competence and autonomy metric, but no relatedness (unless you discuss meta with your community, but that’s not intrinsic to the game).
Competence in Balatro comes from learning what the best combinations of cards and jokers are. You increase competency each time you understand how a card slots into a strategy. And there are a lot of cards and combinations.
You still discover new cards 15 hours into the game. Your next run desire is to test out your ideas for what’s possible: I’ll expand on that later. Tailored challenges are another layer on top of this.
Autonomy-wise the learning phase of the game feels great. You’re incentivized to try new builds by the constant unlocks. These unlocks are tropic to what you haven’t been using: you’re not using Pairs? You unlock the joker that buffs Pairs.
Balatro pushes you to explore new avenues of play and gives you enough control over your deck to make most decisions impactful.
The systems reward going all-in glass cannon style on a strategy.
Flushes? Change all cards to the same suit. Pairs, Three/Four of a Kind? Go for all enhanced Aces. There are so many ways to decide how you play: from the starting deck to the tarots, planets, jokers and more.
I did feel some of my autonomy lapse in later antes (“levels”): you’re constrained to play in a few specific ways, and hit specific build requirements, to win. With random-heavy elements, and no tools to adapt strategy mid-game, you find yourself a loser before knowing it (skill issue?).
The feeling becomes “I had no other choice“.
The prediction machine
Our brain is a prediction machine. Artificial Intelligence and neurology amateurs might be aware of the Reward Prediction Error concept.
In very short: we can view basic behaviour adaptation (i.e. learning) as two signals: the prediction and the reward.
Prediction is the expected reward, information about the future. Reward is, well, the actual reward, the result of our actions. The difference between them is the Reward Prediction Error.
A positive RPE means we got a better reward than expected. That should adjust our behaviour to approach, do more of what we did. It also adjusts the next predicted reward value in the same context (hedonic adaptation much?).
A negative RPE means we got a worse reward than expected. So we avoid or expend less effort for the same behaviour in the future.
Positive RPE = do more of what worked. Negative RPE = do less of what didn’t work. This is oversimplified and RPE isn’t the only learning system, but a simple model to think about in the context of Balatro (and games, and life).
As I mentioned in the competence bit, in Balatro you learn new cards combinations. You fill the competence need by improving at what you do. You improve at what you do by learning what works and what doesn’t. You learn what works and doesn’t through Reward Prediction Error. You get RPE by making predictions and receiving rewards (or not) from those predictions.
So, how do you make predictions in Balatro?
The non-abstract nature of the mechanics lets you imagine the potential effects of your choices. You can straight up calculate the chips a hand gets you. And some people do. But even without that you can produce a rough estimate of what works best for the current context.
The explosive combinations of discrete numbers and card effects creates an analog space you can’t solve without tools, while being simple enough for our pattern matching abilities to shine.
You make a gut-prediction, test it immediately, get the reward, RPE adjusts your model, rinse and repeat. It’s a constant cycle of learning that feels so good.
The predictions you make aren’t just playing the cards, but also building the deck. Whenever you choose a strategy, you predict you’ll get some cards rather than others from the shop; when you get the card you’re expecting… yes!
And when you get a new, even better one… oh boy. Positive RPE through the roof.
It’s the pleasure of building a complex machine piece by piece (deck and jokers), then test how well it works (hit big numbers).
I wanted to write even more about predictions and interesting decisions (what makes an interesting decision?), but this is long already, so that’s for another time.
