Poker: RTA Solutions

A few weeks ago I wrote a post about bots and Real-Time Assistants in poker, and was reminded yesterday that I had more to say about how to prevent Game Theory Optimal solutions in general.

Product Solutions

You may know that I think product solutions are under-used for poker’s problems, and that applies here too.

Burn cards: I started playing online in 2001, and one of the big complaints you heard was that there were no burn cards, like there are in live poker. They’re used to reduce the advantage that marked cards could confer – after preflop betting is complete, someone who can read any card marks would know the door card. By burning it, they can’t make preflop bets accordingly.

Cards can’t be marked online, but we could still rehabilitate the tradition, as long as we give it a twist. If we show the burn card, face up to the table, we damage the GTO solutions that solvers have found. People no longer have the correct ranges or distributions, and instead we reward players for thinking on their feet. You’d need to make ~50 times as many solutions to keep playing perfect GTO, and you’d need to write your own solver to do so. This massively raises the barriers to RTAs. If you like, you can burn & show two cards, raising it to ~2500 times as many solutions required.

I should imagine there’s plenty of room at the top for improving GTO solver efficiency, even if it’s just ASICs (or throwing money at AWS). But as I say in the earlier RTA thread, what matters is making it harder to do it on my site than on yours. Why bother finding 2500x as many solutions for my site if you can just spend your time elsewhere?

Cash drops: RIO have a promotion called splash the pot, where they add a few big blinds to random pots. My favourite thing about it is that it breaks solvers. Your ranges and frequencies are very different if you’re opening into a 2.5bb pot than 1.5bb. If I were to optimise this for breaking GTO solutions, I’d perhaps randomise 0.25bb increments, up to say 2bb. I’d rake a little more, and pay for it like that. You’d need 8x as many GTO solutions, and 8x as many population tendencies to estimate. The latter may not even exist anymore – you’d certainly spend a lot of time waiting for them to settle, as players figure out the best response to being UTG in a 3.5bb pot, and others trying to exploit them.

Rule changes: At Unibet, one of my dreams was to have a version of NLHE where a rule changed every orbit. This time, there’s a fifth suit (or we remove clubs). This time, deuces are wild. This time, you see the bottom five cards of the deck before the deal. This time, a straight beats a flush. And so on and so forth.

This proved impractical to build because it required changes to the deepest parts of the game engine. These are parts of the software that you really don’t want to touch unless you really need to, because bugs can be very bad news, changes can cascade, and regulators need to become involved. Someone would probably need to write their software from scratch to do it, but it obviously messes with GTO solutions. I’d still love to find somewhere to do this though. Think of the promotional possibilities, if nothing else…

And of course, you can combine. If we’re on x2500 as many solutions required by burning two cards, and we then have 8x cash drop variations, we’re up to 20k. That’s a lot of computing power, particularly in an era that Moore’s Law seems to be ending.

Marketing Solutions

If there’s someone in your game with an idiosyncratic play style, your attention should switch away from playing GTO to trying to exploit that style. You therefore break the need for solvers by having more players like that in your games.

I’ve heard this described as a marketing solution, a phrase I don’t particularly like. As you know, I think there are plenty of product solutions to this problem, but as they’re one degree of separation from the topic at hand, I’ll put it into its own category. As anyone who’s played on Unibet since the relaunch in 2014 knows, I heavily favour that site strategy, and this is another factor in its favour.

Technical Solutions

I won’t go into how a site might catch RTAs, all I’ll write publically is in my last post on the topic. But there’s a more general solution: have a 3rd party do it.

Back in ~2008, I bought a computer with a ton of ram, and datamined a decent percentage of all online games at my stakes. This was back when observers were allowed at each table, and you could record the hand histories. Asides from the fact that this was very helpful for my job at the time (I was playing for a living) I was idly thinking about this idea then too.

If I had a way to detect RTAs given enough hand histories, I could make it as a 3rd party. Then I would sign deals with poker sites, who would give me those hand histories. I then detect collusion, bots, RTAs, and money-laundering, and charge for the service. The sites don’t need to employ people to do it themselves, and because I would operate at greater scale than any single operator, I can dedicate greater resources than they can.

You also get greater sample sizes – not of individual players, but of the playerbase as a whole. And you can potentially spot someone cheating on one site, and find the same player on others. You’d never actually have any account information, it’d just be an anonymous player ID, so I don’t think you run into regulatory problems as long as your partner sites mention in their site T&C that they’re sending you hand histories.

There are problems. First is GDPR. Then regulation. Then security concerns. Sites have to trust you (partly this is what I’m thinking about with bot verification). I don’t actually know how to make a solver at the moment. And of course, if everyone has this service, it breaks my idea of security: there is now no less-secure site you can move your bot/etc to, and now you have to concentrate on beating my detection. As there’ll normally be more money in cheating than detecting it, it’s not a great arms race to become involved in for me.

I don’t think any of these problems are fatal, but they’re definitely a concern. That’s why I haven’t done it, plus of course, my network of people working in this area isn’t as strong as it could be. It’d be pretty fun to try though.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s