A long, preliminary discussion on how I determined my first test subject in the 3/6 Project

Initially, I had planned on programming strategies from the books Winning Low Limit Hold'Em by Lee Jones and from Small Stakes Hold'Em by Sklansky, Malmuth and Miller.

However, while working with the Lee Jones strategy, I realized while programming the preflop strategies would be fairly easy (if not time consuming), programming the postflop strategies were not as easy. Postflop play is very nuanced. Some flops, like top set, a flopped straight or better, or top pair top kicker with no discernible draws can be fairly easy to play. However, the strength of straight and flush draws can vary greatly. As SSHE puts it, some gutshot straight draws can be fairly strong, while some open ended straight draws can be fairly weak.

Even if I possess a perfect understanding of the postflop strategic considerations in each book, I can err in programming the strategy into the simulator's complex interface. One incorrect strategem given for one particular drawing situation, and I could leave the simulator making multiple, repeated postflop mistakes.

Wilson Software took great care to program their advisor and more solid profiles to play solid postflop poker, even if their preflop considerations are a bit loose. I made an initial attempt to program preflop strategies for the above two materials, along with a simplified, more passive version of their postflop considerations. For example:

1) Stronger straight and flush draws were usually bet aggressively on the flop, but played passively on the turn if not hit.
2) Low end straight draws and weaker flush draws were played more passively.
3) Top pair was played 50/50 aggressive and passive unless we held top or 2nd kicker, and even then, TPTK was played more passively with any strong straight or flush draws on board, especially if the board presented possibility of a made straight or flush.

To properly simulate game conditions, I initially set the rake at 10% $3 max with a $1 jackpot pot, a $1 toke after each pot won, with a minimum $13 pot required to toke (leaving the winner with 2 BB), and a $20 minimum to take the jackpot drop. However, the simulator did not allow me to vary the size of the jackpot drop, as is standard at most local cardrooms (most will take the 2nd jackpot drop dollar when the pot reaches $40), so to compensate, I set the simulator to take a 10% $4 max rake with a $1 jackpot drop, which takes basically the same amount from the pot. I also had the blinds chop if the pot was folded around to them, typical in many local cardrooms.

From there, I ran lengthy simulations for each strategy against a typical low limit lineup at a 9 handed table, a mix of 1-2 loose aggressive players, 3-4 loose players and a smaller subset of tighter players. The composition of this lineup varied, and I set the simulator to vary the styles of the other players on different streets to mix up play anyway, while taking care not to have the simulator 'play back', or adapt their strategies to how my test player played.

This strategy sometimes turned a slight profit in the long run, about half a big bet an hour, over simulations ranging from 200,000 hands to 1,000,000 hands. But typically, simulations for both strategies broke even or posted a slight loss. It is quite likely these steatgies struggled due to a limited and, in some cases, crippled postflop strategy.

Still, I performed simulation trials of each starting hand in each position: under the gun, early position, middle position, late position, and the small and big blinds. From here, I took each recommended starting hand and ran it in each position over 25,000 trials to determine if it was profitable from the position stated as played, or if it was a leak hand: a hand that turned a long term loss. Hands like Axs and KQo showed a loss in middle position. Hands like Q10o, J10o, T9s and 98s showed a loss in late position.

Following removal of the leak hands, I re-ran the revised strategy over a trial of 500,000 hands. The strategy, in turn, still only turned a slight profit, no greater than $0.09 per hand (less than half a big bet per hour) against the given lineup.

I then ran a 500,000 hand trial using the Conan the Librarian profile from the simulator, a typical tight aggressive profile (though with a relatively loose preflop strategy) offered as an advisor. This profile turned a consistent $0.20 per hand profit, about one big bet per hour.

1) I decided at this moment to abandon programming a strategy from scratch, and instead focus on exploring the existing advisor strategies, given they had more solid, fleshed out postflop strategies (and evidently were working). I figured that, once I refined the best advisor strategies on the preflop level, I could then explore the postflop strategies, see what makes them work, how well they reflect the strategies taught in top materials like SSHE, and develop a custom strategy from there.

2) While I like that the given profile can turn a profit, my ultimate goal is to devise a strategy that can crush 3/6, to the point of potentially earning 2-3 big bets per hour. A player using such a strategy could grind out a part time income at the local tables. I wanted to focus on improving the given strategy if possible.

3) This strategy turned one big bet per hour despite loose starting hand requirements that played hands like J10s and 77 in early position. I wanted to key on leak hands and see if I could improve the win rate by eliminating the leak hands while retaining the postflop strategy, and maybe even improving the preflop strategy on playable hands to maximize winrate.

4) One useful inclusion in this software are the duplication of profiles with subtle differences. A profile called Gypsy Rose plays the same hands as the Conan profile, and plays the same postflop, except much more passively preflop, limping everything outside of AA and KK in early position for example. This will allow me to cross-test how different hands play when they are open limped and when they are open raised.

Conventional wisdom states that you should always enter a pot with a raise when 1st in. However, conventional wisdom also states that certain hands play better in multiway pots, such as Axs, small pairs and suited connectors. I will look to run trials of hands where they are both raised and limped, from under the gun, early position, middle position, late position, the button, the small blind and the big blind.

Apologies for the length of the post, but I wanted to explain the preliminary testing that led to my decisions on the parameters of my forthcoming tests, which will initially focus on finding expected value of given starting hands given a control postflop strategy, and determine which hands are profitable in each position, and whether they are most profitable when raised or limped.