Python Playtesting: Crafting the Perfect Board Game - Alla Barbalat - 2026

So, we've had a couple of pretty heavy serious type talks coming up immediately before this. Time for a bit of a bit of a change of pace. This talk is about play testing with Python and creating board games. Please welcome Ala. >> [applause] >> Hi everyone. Thank you for that intro. Very excited to be here today and talk about Python play testing crafting the perfect board game with Python. Really excited to welcome you all here to my talk today. Thanks for joining me. Before I get started, I would love to get a sense of how many folks here enjoy playing board games. Good, that's probably like the majority. And then I'm curious about how many of you have either made or designed a board game? A good handful of people. Wow, that's impressive. That's more than I would have expected. So, that's good to know that maybe I could get some good feedback from folks after this. And I'm also curious of those folks that have made or created board games, have you simulated them or tested them with Python? All right, just one. So, be a really wonderful to chat with you after. Awesome. And with that I'd love to get started by introducing myself. I am a former lawyer and I feel like starting my professional career as a lawyer has really shaped how I view the world. So, I included that. I also have a really crazy shepherd husky mix. So, she's going to be a little bit in this presentation too. So, I hope you all are prepared for that. And then I'm on a somewhat of a career hiatus at the moment. And during this hiatus I restarted the San Francisco chapter of PyLadies. >> [applause] >> Thank you. And I bring that up because one of my goals as the current lead organizer for PyLadies San Francisco is not just it to like of course improve the Python community and share information and you know, enjoy each other's company. But one of the things I really love doing is creating a platform for new voices and amplifying voices that might not get heard otherwise. So, I really love finding speakers that maybe have never given a talk before, maybe are really nervous about giving a talk or maybe feel like, oh, I have a personal project, but make that into a talk? So, something that I've really been working on is kind of inspiring those folks and giving them the confidence to present whether it be at the PyLadies San Francisco meetup or at a conference or another space. And I hope that this talk in some way also serves as inspiration for that. Also, I put this in italics. I guess I'm also a board game creator. I don't really think of myself as that. Thank you. This is the first board game I've ever created. I think of myself more as someone who enjoys playing board games. I really love social games, cooperative games, and competitive games in the right environment with the right group of friends. So, definitely maybe a little more interested in social and co-op games. So, with that I'll walk you through a brief road map of what I'll be talking about today. First, we'll discuss building a board game prototype in Python for fun to make it fun. Using simulations to test the game logic. And then looking at what worked, what could have been better, and what comes next. So, the idea for this game was like very random. Like I said, not really a board game designer. I've had a lot of different hats that I've worn professionally, but none in the game space. But one day I was standing in line at a restaurant waiting for the bathroom and I was struck with a moment of inspiration and I thought, wouldn't it be cool if I created this board game and it really had two ideas that really sparked my interest. The first was to create a game that had a very dark satirical political statement as the basis of the game. So, I'll get into that a little bit more, but not enough to hopefully make people feel uncomfortable. And then the other idea was to have a board game that starts out as cooperative and then at some point in the game there's this chaos mode that's triggered and the game becomes competitive. And my hope was to have a variety of different competitive conditions that could be created. So, different win loss conditions based on whatever chaos mode was entered. And unfortunately with this prototype, I didn't actually include the chaos modes yet. Though very excited to work on those eventually. And So, as this process was going, I got inspired. Never really thought I would take the inspiration to heart, but somehow managed to put the initial ideas into a Google spreadsheet which I don't know if it was my lawyer brain or something else. I was like, oh, a spreadsheet. That's like natural place to put a design for a board game. So, started there and then built out a lot of the game and then made my partner play some rounds with me. And even outside of the fact that obviously there was no UX UI because we're playing from a spreadsheet, there were other things missing and lacking in the game logic that I saw needed improvement. And even though it wasn't especially fun, my partner is still with me. So, that's good. >> [gasps] >> And I was inspired to take that spreadsheet and then put it into a Python program with the goal of then simulating the game a bunch of times to figure out how could I improve the game logic? And I'll just make a brief note that unfortunately because I started with the spreadsheet, I was like, oh, it'd be really easy just to like get a bunch of CSVs and import the game data that way which definitely got me up and running really fast. But I feel like if I had to start over from the beginning, I would not be importing the game information via CSVs. So, that was a whole weird process that I went through that now I might have to change eventually. But I do have a terminal based version of my board game. And as you can see there is a very, let's say, compared to other board games, kind of boring map here. So, the initial idea here was that there would be different spaces that the characters can move through. And I'll just give you kind of a very brief overview of what the game entails. So, basically, like I said, it's very dark satirical take on let's say a political system. So, in this game >> [laughter] >> there is this unstable monarchy that is trying to accumulate wealth and their actions have consequences and that affects the townsfolk. And depending on the consequences, it can destabilize the kingdom and lead to a loss condition in the game. And the way that the game operates is every player is a character. I have three characters so far. You can see them on the slides and every character has its own character deck and they play three cards of a five card hand for each turn. And then at the end of everyone's turn at the end of the round, an event deck card is drawn. So, that's just like the basic game mechanics of this game. And once I put this together, then next step was to figure out, okay, I have this terminal based Python program. Now I have to run these simulations to improve the game logic to make the game fun. But what is fun? So, that was a big question to answer. A little bit like more of a big philosophical question than I think I can answer in this talk. Though I could try. I feel like I might get removed from the stage. In any case, so I needed to pare down this question of what is fun to specifically focus on what is fun in terms of the specific board game that I'm trying to make. And so, what's great is that with Python, I was able to ask this question in a much smaller scale for very small pieces of the game so that I could build iteratively in the game logic to figure out what would be fun in terms of the game logic for this specific game. And while I was working on that, I figured this is kind of the underlying theme for how I'm going to run my simulations and how I'm going to figure out if the game is fun. So, before I did any simulations, I put together this expected criteria for evaluation. And a lot of it had to do with like the length of the game. Is the game too long? If it's too long, people get bored. And what are like the lengths of the rounds? If the rounds are too long, then people are going to get bored while like someone else is playing. And how do I figure out the movements? That was definitely an issue in the spreadsheet version was moving around with felt a little bit unpurposeful. So, I had to basically work on that in the revised version of the game. So, those are like relatively quantitative. On the more qualitative side, when it's to figure out like what are understandable and achievable goals for the players in the game. I also wanted to create a balance of strategy and luck. So, if it's impossible to win the game, then that's not going to be a lot of fun. And if you can like randomly do anything and win the game, that's also not fun. So, figuring out how much strategy do you need, how much luck do you need, and is it possible to win the game? Which is very important for a cooperative games. With competitive games, to a certain extent like the level of difficulty in the game doesn't matter as much. What matters is the specific competitors that are playing the game. So, you can have a really easy game like tic-tac-toe, and two people that have very good at are very good at tic-tac-toe, and they're still enjoying the game even though it's very simple. But, if you have a really simple cooperative game, then it's not going to be that fun if you could just easily win the game. In the same vein, if it's too hard and you can never win, you're not going to invest in the game or want to play. So, those were kind of the initial ideas. I also wanted to look for replayability. That's also a little bit hard to determine. So, these are like more of the subjective criteria that I started out with. Then, I actually started testing. So, I used three different methodologies. And so, these are all like Python-based simulations that I ran. The first was a fairly like rudimentary series of functions that I ran that were maximization-focused with breadth-first search for strategic movements. So, the strategic movement was limited only for the specific turn in question, and it was specifically only to maximize the player's wealth or the townsfolk population. So, there was other things that you could maximize that this function did not consider. And then, there were other trade-offs and costs that were not considered. So, very like rudimentary, very basic strategy for playing the game. Then, the second methodology I used was just a random action function. As I mentioned before, it's really important to have a game that's winnable through strategy, especially for a cooperative game. So, the idea here is you take this random action function and then see, "Well, if I just do random things, can I win the game? And how quickly would I win or lose?" So, that was a very uh interesting methodology to use in this particular game. And then, I also used a series of LLM-based AI bots with varying amounts of context on the game to see how they would perform and get additional insights. So, with that in mind, here's just an example of some code that I actually used in my simulation. This is for the randomness. So, basically you can see that you are basically doing like the most random thing possible, and then you would just do that over and over again for every single player's turn and see what happens. So, what was interesting is I actually thought with the random action that it would never win. So, I was really surprised to see that with random action, this game actually was won twice. And these specific graphs that I'm showing are from simulations I ran last night. So, I ran 100 simulations on random action and 100 simulations with the very rudimentary strategy of just maximizing wealth, maximizing townsfolk, moving strategically per turn with breadth-first search, and then not considering other costs and uh features of the game. So, the random action, two of 100 wins. I mean, obviously that's not a lot, but previously when I had run the random, I had never gotten any wins. So, two in 100 still was like extremely surprising to me. Then, there were 58 losses. That That made sense. And then, there were 40 that never reached a resolution. And so, uh I kind of thought that I had just underdeveloped the game because I only included 17 event deck cards. So, after 17 rounds, the game ends because there are no more event deck cards. And I was like, "Oh, I need to like add more." But, actually this ended up being a very good feature for this prototype because it prevented the game from going on too long. Uh especially like later when I talk about like the LLM-based bots, like it would have been like if I had 100 event cards and they just kept playing, it'd be like way too many API calls for me. So, this ended up being perfect. And I do actually think over like even 17 rounds is too long for this game. So, it was really good to see that with the random action, not only are you like probably going to lose. Like otherwise, the game also just lingers. And so, the lack of resolution was also to me a good sign because the game is really about trade-offs. And if you're not using any strategy, you're not considering the trade-offs that are built into the game. So, when you take an action, there's a lot of consequent actions that happen as a result that would just never happen if you're playing randomly. So, with this very rudimentary function that I mentioned that basically maximizes wealth, maximizes townsfolk population, doesn't maximize anything else, and then strategically moves through the board per turn, I was able to increase from two to 12. And I was actually surprised by this. I thought there was going to be a higher win rate here. So, this was interesting. I mean, it's probably overall good because there's a lot more strategy and a lot more consideration of trade-offs that should be made in the game. So, uh this was a good result. And then, obviously there's also a lot fewer games that hang. So, this was also a really good result because that means that if you're actually using some strategy, you're more likely to end the game, which is a good result. You don't want to play a game that has no resolution. It's more satisfying to get a resolution even if you lose than to have no resolution at all. So, this all was like looking very good. And then, I also last night ran 20 simulations with the LLM-based bot. And this is like the most sophisticated bot that I had playing the game. And they're just making API calls to uh an LLM, and they won about half the time. So, I feel like this is pretty good. Like this seems like reasonable. If the bot's like running about half the time, has a lot of context on the game. So, I felt pretty satisfied with that. And so, after I'd run all these simulations, I actually Well, not just these simulations, but I ran like a bunch more that aren't on the slide. But, during this whole process of running these simulations, I ended up using very different criteria to evaluate the simulations than I mentioned previously. So, unfortunately, I didn't really get all that like time-based data because the simulations I'm running directly in Python that aren't based on the LLM run instantaneously. And then, the LLM-based ones often times actually took way longer than expected, especially the ones that were like getting a lot of context, but still wouldn't be in any way like matchable to a human being. So, uh before I created my more sophisticated LLM bots, I was having a trouble getting like win results at all. So, I actually weakened the win condition, and then I was able to achieve a win with the LLM bots. So, then I went back, and I was like, "Well, what's going on here? Like, I need to actually sit down and just play the game." So, I took a step back to some human in the loop, and I played a three-person cooperative game by myself in the Python terminal. And took actually a really long time to play 10 rounds, like surprisingly long. But, I was able to win the game. I do know every single card and like have a sense of like what could happen at any moment in the game. So, that could have also helped. But, uh from that, I figured about like 10 rounds max for like a three-person game would be good. And I addressed the criteria that I used to evaluate the simulations. I also then went back to the original win condition and adapted the LLM bot so that I could get better performance from them. So, uh basically, the actual criteria I used were like the number of rounds. So, like looking at the 17 round max, like did the game end or not? The variance in game length. So, with the random function, I was able to gain like uh lose the game within like three rounds, which was really interesting. So, kind of looking at how many rounds were there total. Looking at end state conditions, so like the proximity to the win condition if the game was lost or if the game was just hanging. And then, uh I also ended up looking at these other things that I didn't really consider as much before. Like handling impossible hands was really important. And also looking at the AI reasoning for the LLM bot decisions and their asks. So, and then obviously, as I've already discussed, the effects of random action and the effects of rudimentary strategy. So, with that, did I craft the perfect board game? I'm sure it's pretty obvious the answer is no. But, I feel like I accomplished my goal because I now have a prototype that I feel comfortable playing with my friends without having to worry that I won't have friends after. So, yeah, goal accomplished. So, this was a really fun project and I feel like I learned a ton about game design and how you can use Python to simulate. And with that, I have a lot of like possible next steps for myself that I'm really excited about. Definitely want to actually play the game with friends. That would be really cool. Hopefully with the physical version that it maybe I'll just have one prototype version that I create. Uh definitely want to go back and add the competitive chaos modes that I mentioned. I want to build out more features and characters in the game. And I really want to look at more complex logic for the max functions. Like I said, they don't really look at trade-offs fully, so that's definitely something that I could incorporate into the game. Also want to look at potentially doing machine learning algorithm-based AI bots. So, kind of the old-school AI and that would work for both like cooperative and competitive modes. And then potentially for future competitive modes, I could do mini max or game engines. So, like a lot of future possible simulations that I could be running, as well as money Carlo simulations. And that So, hopefully the key takeaways from this talk are to take a chance on creative projects. And sometimes when you have this creative project, you can use Python to take a creative question and turn it into a testable one. I will say, sometimes you have a project on the back burner or you just don't feel confident working on it or you're not sure about your talk proposal. And I really hope that this talk inspires you to go on and do that. Often times, I very begrudgingly take my dog out into the rain. And then we get to walk around and see beautiful rainbows together. So, uh hope for many rainbows on your Python path. And happy to answer any questions or connect on LinkedIn. >> [applause]