Writing Super-Turbo bots for fun and profit
Cliff’s Notes:
Beat: It got busted.
Brag: I wrote the poker brain for DeepComputer/TheFrenkel
Variance: I never used the thing anyway and never could have because lol($3.40)donkaments.
tl;dr version
Background
Zoo thread about people getting their accounts locked
Cast- The leader… let’s call him Josh. Josh lives in Israel and has not much but time on his hands. He’s a smart guy but so far a relatively unsuccessful poker player.
- Me: I’m the poker player, because someone has to know something about playing SNGs. I’m also the teacher.
- Other people I don’t know.
Timeline
August 2007
Josh approached me for SNG coaching as most people do, through a PM on 2+2 asking what I charge and what exactly I do for coaching. I gave him my standard reply asking about his previous experience and ambitions for his SNG game.
Quote:
I have been reading some of your post lately. Really liked your review on the new SnG book by Collin Moshman. I started to play not long ago, I use the SngWiz, read all I can find, watched lots of the sngicons videos. I did a couching session with Collin, but I don't feel it help me a lot.
I was wondering if you have the time and energy to try and help me out?
After a few emails, he decided he wanted to get to where he could 8-table the $16-turbo level on Stars. Here I learn that “TheFrenkel” is his first account on Full Tilt and “DeepComputer” was a second one he created for rakeback. He seemed very serious about improving as poker offered a significant improvement to his financial prospects. Playing that level of SNG is fairly brainless and a good way to print some money, so off we went.
September-October 2007
Josh continued as a normal student. He seemed dedicated but suffered from the bankroll problems most people do: taking money out for expenses and not being able to move up. He asked the usual questions and I gave him the usual answers. Most of my coaching is very ICM-mathematics based. This is because SNGS are a solved game, and whoever makes the fewest mistakes wins the money. He paid me my usual fee for my coaching. He was definitely good enough to have accomplished his initial goal of 8-tabling the $16-turbo SNGs for a profit.
November 2007
Josh suggested writing a real-time advisor program to help him not make mistakes. I told him explicitly this was a T&C violation and it would be risky. I also told him the same thing I tell everyone who starts pushing me that way: “I do not write code.” He said that was fine, that he had a team of code-people ready for that, one of whom was a first-year PhD mathematics student.
December 2007 – January 2008
Josh got back to me with what he claimed was a working version of his real-time advisor thing. I didn’t really care because it is of no practical use to me. He then sent me the strategy guide he gave the programmers. This was something I could understand. He asked me to help him with it and agreed to take on his “student” on a contract basis. I knew he was writing a real-time advisor and it was my job to develop the strategy to make it actually win. This is not very hard given the nature of SNGs
if you understand them.
By the middle of January, I was looking through BASIC code and finding the leaks in this thing’s game. We’ll call it DeepComputer for clarity, though that name is pretty lame to give to a poker bot for obvious reasons. The rest of the people involved were doing something like playing $23 turbos on FT and Stars. Josh was fully apprised of the risks he was taking and I could give a care about the moral implications of “cheating” in a $2 SNG. The greater challenge is actually making this thing work. Once it did, Josh and the others were off on their own and I got paid my “coaching” fee. We agreed to continue it on an “ongoing” basis, but not much happened for awhile because apparently you can’t make much money at the $2 SNGs. Who knew?
As a sidenote, Josh had offered me a copy of the program in lieu of cash payment and I had refused on the grounds that it’s completely useless to me. What good is risking getting my funds confiscated for the joy of cutting my hourly rate by a factor of three? SNGs may be solved, but it’s still not easy to beat anything above the micro-stakes without some amount of human intervention.
February – May 2008
A lot of stuff gets done to DeepComputer. I revise a lot of the strategy. At this point, it goes from scraping results from various outside programs to being a self-contained Nash Equilibrium-based calculation engine. It was my idea to make it a stand-alone program because there a lot of important factors that are not accounted for in off-the-shelf IMC-based analysis software.
May 2008
Super-Turbos! The super-turbo structure starts with 10 BBs and lets the fun go from there. Initially, I didn’t think these were beatable, but that turned out not to be true. A number of good SNG players expressed concerns this structure was was a slam-dunk for botting. No one important listened. Josh contracted me for some strategic tweaks in the program, which I gave him. He and the others went off to do their thing and I went back to whatever it is that I do.
July 2008
The DeepComputer account wins the low-limit SNG leaderboard. Normally, I wouldn’t give these things much notice except that Josh was behind on his payment for the work I had done earlier. I’m fairly easy-going about students paying me. If they don’t make money, that’s probably my fault anyway and I don’t expect to get paid for not producing results. It is mostly just for my own entertainment anyway, but if I see someone is making money and owes me money, I expect to get paid. I get a little bit pissy with Josh and tell him I’m considering ratting him out for fun if he doesn’t pay me. He finally pays me. I consider the matter settled and go back to whatever it is that I do. DeepComputer/TheFrenkel beats the super-turbos for about 7%, which I think is solid.
August 2007
DeepComputer/TheFrenkel gets busted. The primary reason for this is some really stupid multi-accounting/play patterns that triggered some sort of Full Tilt investigation. I doubt they went in looking for bots but that’s what they found. There were some other lazy/careless things on the programming side that would have stuck out as well. The bad thing is that a lot of innocent people got their accounts frozen in the process. Most of the successful super-turbo players are just very solid push/fold SNG players willing to put up with a low ROI for a huge number of tables/hr. To those of you caught up in Full Tilt’s clumsy dragnet, I’m sorry. If I’d known the investigation was going to suck that badly, I would have cut it short with all the info I knew at the time.
Figures
FAQ
Q: Did you ever use DeepComputer?
A: No.
Q: Why not?
A: It doesn’t beat the games I play, and my hourly rate is 4x higher in those, so me actually using it would be pointless and counter-productive.
Q: In one sentence, why did you help write it?
A: I wanted to see if it would work.
Q: How does it work, exactly?
A: I’m not going to tell. The only thing I’ll say in that regard is that it was much simpler than I thought it needed to be to win at that rate.
Q: How much money was in the accounts?
A: I’m not sure. My guess is between DeepComputer and TheFrenkel maybe about $5k.
Q: Can I have a copy?
A: No. I don’t have a copy of the actual software and all the development information I generated is proprietary. I use the same stuff to teach real players.
Q: Are the players still using it?
A: I don’t know, I don’t really want to know, and I doubt it based on how badly they botched the account management aspect last time. Anything’s possible I suppose.
Q: You’re an a**h***.
A: That’s not a question, Ambassador. I don’t really care if people have a different opinion on the ethical implications of breaking the T&Cs of an online poker site.
Q: Why did it suck heads-up?
A: I guess because the apple never falls far from the tree? I don’t know. I’m not even convinced it did suck and that finish distribution wasn’t random or an artifact of playing correctly on the bubble.