Polymarket Weather Trading Bot Built With Claude Code (Full Strategy)

Added: 2026-05-27

[00:00:00]So, I was on TikTok at midnight doing what any responsible adult does, and I found this guy selling weather trading tips for Polymarket, $25 a month. I built a model that trades the weather on Polymarket in Cal sheet, and it makes me tens of thousands of dollars a I paid for it. So, now you can use it, too. And it did absolutely nothing. 25 bucks gone. Thanks, mate. So, I did what any normal person would do. I downloaded his video, gave it to Claude, and said, "Build me a better one." Now, I should mention that I didn't code a single thing, not one. But, it ended up getting me results like this, and this, and this, and this, and this, I think you get the idea. But, to get to those results, I kind of have to rewind to about a week ago.

[00:00:42]So, back to the beginning. I asked Claude that I wanted to make a bot and what I wanted to do. So, I told it that I wanted to replicate this video to see if it was even possible. Wasn't even sure if it was. So, simple as this.

[00:00:54]Add files, downloads, and then Optical's video, which you just saw then.

[00:01:00]And then, boom, just let Claude do its thing. It did need to kind of run through, analyze a few things, but I just sat there, had a drink, and let it run through. So, what is the TLDR? Well, basically, it got 173 different organizations that do temperature reading, and they call this the ensemble. So, 173 members the ensemble, and here is what they're all a part of. And what they do is they take the average of all of those, well, the mean of all of those, and put it together to get one singular temperature, and then we base our trades off that. But, it's not just that. We then also back test a little bit, and then also do some test going forward to see if that mean always leans hotter, leans cooler, or right in the middle to then make our educated guesses.

[00:01:40]>> [music] >> Basically, you put this data into a chart, and here's what we get. So, we get the ensemble says that it's going to be 56.7 in NYC, but we've noticed that there is about a degree to a degree and a half between the actual the temperature that is executed and the temperature that they think. So, here is the real temperature that we're thinking about with the 173 members, and this is the range. Now, what does all this gibberish mean? It basically means that we can see where the prediction of the market is. We can see where the historical testing of it is, and then we can pick a bracket that works for us.

[00:02:13]And if you pick too close to the bracket, even though you might be right, you can also be wrong and then lose a bunch of money. So, with our first trade, we actually ended up losing $24 because we were so close, and we weren't taking the money outside of that risk zone of the bracket. We like to have a 3% or 4% difference to actually have a chance of winning. So, where do all these numbers make us land? Well, basically, the real bias, what they think and what we've corrected for is 55.3, and the ensemble is 56. So, we want to stay out of this range, which is the 55 and the 56 range. When we head over and have a look, we can see the 54 to 55 range has ticked no. So, it's actually gone higher than that right now. So, it means it's quite a warm day. We can see the 56 to 57 rate um temperature is at a 71% chance, so it's up. So, you can actually see the 58 to 59 is a safer bet if you're going to kind of look at it because what we can see here is the ensemble says that it's going to top out 56 to 56.7. So, just say if we go 57, the 57 range still hits on this top one here. So, if we take the 58 to 59, we're in a good spot. So, that is a 25% gain that we could end up taking a swing on.

[00:03:22]Now, a quick note here is that we've actually found the no being better than the yes because only one yes can win, and the rest will win Well, the nos will win. So, the idea here is if the market is saying that a certain yes is going to win, or the ensemble is saying that a certain yes is going to win, and you've actually feel like you can get a bit of an edge on a no with a sizable buffer away from where the um ensemble is predicting, that's when you're actually going to end up with good money because if it overshoots or undershoots by even 2°, you can still win. When you take the yes, you need to be spot on and even though your ensemble might be right, it can just go half a degree or a quarter of a degree over and then your yes gets wiped out. Whereas your no, you can go further up the risk curve, do a safer bet and then end up actually winning and stacking those gains. So that's just something we found. So you notice a lot of these talkings here, I'm actually coming at it with the mindset of taking a no, not the yes. And you can see here in the bot what we've actually done is the buffer range that we have here, 0.7 degrees, no good. That's why there is a caution sign here. And even the 58 57 at 2.7, we want to have a bigger buffer over that 3%. We can see here the edge is only 6.5. And the thing is I've been trading this bot, we've been testing it out, playing with it. Honest take today, skip NYC. And this is what we've done over time after time. After we built it, put in some trades, made that initial loss, revisited it, come back, try it again, try it again, try it again. And we can actually see that through the wins.

[00:04:50]We've got uh a 163% win, 64% gain, 19%, 87, 57, 54, 53. We've made some solid gains and some solid wins, but we've put on very specific um price targets, very specific trades even when we shouldn't have to kind of stress test the model to end up in a better position and overall end up in a better spot to end up with predictions. Now I'm just going to check this right now. Can we check the Miami market to see what's happening? So in real time you can see how the bot works.

[00:05:21]>> [snorts] >> It's going to run through. It's going to pull all of the data from the ensemble.

[00:05:26]It's then going to put that data in, check it with what we have done historically, see if there's any edges, check the markets, everything in between.

[00:05:37]So we can see here right now um we've got the different ratios here and I like it likes to give me a breakdown. So skip Miami, 3 days in a row we've had this same 84 to 85 yes signal that has appeared in three days in a row, it's been 80 to 82, meaning the Miami me mean is only six to seven off the afternoon high, which is completely normal there.

[00:06:00]The bias correction that we built for the NYC um is just wrong for Miami and it's not right. And the ensemble is running three to four percent cooler because it's a coastal city and not the 1.4 that we built in. Today the play is NYC if anything, but as we saw NYC signals are all over the place at the moment, so not quite good. Now this is where I ended up having an issue with optical is the fact that while there weren't calls every day, I wasn't sure if the bot was actually working and if my money had just been taken. With this bot, even though the plays aren't there, it's telling me why the plays aren't there and that's why I ended up building this myself. When I was waiting for the signals to come through, I needed to be online at my 12:01 a.m.

[00:06:43]to see if the signal populated, which was quite silly of an idea for it to never happen and I think out of an entire month, I think I got one or two triggers and it did it did nothing for me. So with this, at least I was able to build it. I'm able to ask questions. We're able to build it and manipulate it around what we're doing to end up with a result and end up with a positive P&L.

[00:07:04]And with Chicago here, we can see that they're saying skip today across all um because here's the pattern.

[00:07:10]The range is too wide and the ensemble is too split.

[00:07:15]So that's how the bot works, but how would you actually make it? Well, you could upload the optical video, you could upload this video, take a transcript, put it out there, upload this Claude code and it can just work it and just build it itself, totally fine. It may cost you a bit in tokens. Or what I'll actually do is down below, I'll put the skills doc. So I'll get Claude to create a document that you can then upload and then Claude will then know what this has been built and then give it to you so you can then play it there.

[00:07:42]Now, I will say this bot isn't 100% guaranteed. Here we can have a look at the P&L win rates, all of that. We can see our win rate is 58.8% 17 signals and our total staked return like you can kind of see it here. And you can see it's not just straight up, it is up and down, it is bumpy. And we can see all of the trades that we've made, how they've gone, and if we've won.

[00:08:03]So, if you do take this on, um not financial advice, do your own research, all of that kind of stuff, you're not going to this isn't just going to print money. It is going to give you the edge and it is going to give you an ability above everything else just to kind of have a chance at winning. Um being all going to guaranteed win, you can't just set and forget this and let it run. Now, what is the strength of this overall system? It may not be the daily trade, but it's finding those trades that are just completely anomalous and then being able to full port it. Now, basically, I want to I need to give you this example.

[00:08:34]Chicago on the 21st of May. It says the ensemble was saying that it's going to be 57.

[00:08:41]Uh four. Now, 76% of the 173 members were saying that it's going to be less than 59% and basically, you could buy the yes, which was the below the amount and there was a 22% edge. And then what I was able to do is enter with $55 at 61% and then end up with $90 out. So, I think it was a $35 gain. And the idea here is being able to pull the trigger and move in with size with a little bit more conviction on something backed by data that you've been able to back test.

[00:09:13]The idea here is we had a 70 we had 76% of saying it, but there was a 60% chance that the market was saying it, so there was that big discrepancy, so you were able to take that swing and make that money. So, I guess the question is where am I sitting at overall? Over the last month, I'm up $10, which doesn't sound like much, but it is around sort of 15% of the portfolio. Saying that, the very first trade we were down $24. So, from minus $24 to plus 10, up $35, which is around a 50% gain on the actual staked amount. So, yeah, right now the bot is working. The bot's cooking. I wanted to just put this out here. I want to make this video to show you what is happening and what you can do with Polymarket, with something like Claude, with some AI to pull it all together to actually make some wins. Because looking at these wins, these wins are absolutely juicy.

[00:10:04]They're fun. All that good stuff. So, yeah, hopefully you enjoyed the video.

[00:10:07]Hopefully you're able to find some value in If you did, go down there, drop some love. Also, if you do want to give it a go, download the skills um doc, punch it into Claude, and then you can even paper trade dry trade and then let let me know how it goes on that. Um drop all the comments below, all that good stuff. Had a pleasure making this video.

[00:10:26]Hopefully you enjoyed watching it, and I look forward to seeing you in the next one.