Re-evaluating the testing process

Pardon the absence during a busy time for me. Normal posting will resume shortly.

I have run multiple test trials over the last few weeks, lately gravitating to more specific scenarios in an attempt to more accurately gauge where hands make their money using a conventionally solid postflop strategy. However, I'm running into a problem: I've so expanded the scope of this project as a result that to ultimately complete the task strikes me as overwhelming. Each hand in each situation requires a minimum of 50,000 trials to minimize the variance so we get a useful net profit number.

If I run too few trials, the variance ends up so high that the data is useless. The variance over 25,000 trials can get into the $0.30 range, rendering most trials useless. You need at least 50,000 to get variance within $0.15 and you would prefer 100,000. If I don't run trials by position, an overall number for all positions of a given hand doesn't tell me very much about the hand's profitability (is it profitable when open raised but not when calling other players? Does it profit against multiple players? Is it profitable in late position?). A single number over a large span of cumulative trials won't educate me about how or where a hand is best played before the flop.

Also, we eventually need to transition to evaluating postflop play to see if we can improve on conventionally solid strategy, and there are so many variables there (Yes, I do have a framework from which I will run tests). But one step at a time: if we can simply get a reliable barometer on profitable preflop hands, that is a big step forward.

To make the task less overwhelming, I now plan to do these tests by the sets listed. I will start with pocket pairs, then the suited broadways, the offsuit broadways, the suited Aces, Kings, Queens, then Jacks, then the suited connectors, and then I shall add other hands as low-end hands show the potential for profitable play. I previously hadn't tested the top hands like AA and AK, but I will now include them to provide general barometers of how hands improve in profitability with position.

I will test all groups in the following circumstances:

Open (1st in) call, Open (1st in) raise, limp behind, raise limpers, call raise, reraise

I will test all the above circumstances for the following positions:

UTG, EP, MP, LP, Button, SB, BB

Already, we face a daunting task. Now, the groups:

(AA, KK, QQ) JJ, TT, 99, 88, 77, 66, 55, 44, 33, 22

AKs, AQs, AJs, A10s, KQs, KJs, K10s, QJs, Q10s, J10s

AKo, AQo, AJo, A10o, KQo, KJo, K10o, QJo, Q10o, J10o

A9s, A8s, A7s, A6s, A5s, A4s, A3s, A2s

K9s, K8s, K7s, K6s, K5s, K4s, K3s, K2s

Q9s, Q8s, Q7s, Q6s, Q5s, Q4s, Q3s, Q2s

J9s, J8s, J7s, J6s, J5s, J4s, J3s, J2s

T9s, 98s, 87s, 76s, 65s, 54s

For each group I will do the following:

1. Program the simulator to run trials for each respective hand,
setting button positions for each hand.

2. Run 1st in trials for calls and raises (26 50K trials for each position)
- Button
- LP
- MP
- EP
- UTG

3. Run limper trials for calls and raises (26 50K trials for each position)
- SB
- Button
- LP
- MP
- EP

4. Run raiser trials for calls and raises (26 50K trials for each position)
- BB
- SB
- Button
- LP
- MP
- EP

5. Rip my hair out.

Now, to wit: the simulator can run about 3.3 million hands an hour, give or take depending on what scenario you research. Running 5 million trials takes about an hour and a half. 13 separate 100K trials adds up to 1.3 million trials. Run any of the above listed scenarios for every position and you get either 6.5 million trials or 7.8 million trials. That's about 2 hours for each. Not including time to manually notate the results, that's 6 hours to run trials on one set of hands. And that assumes we can set the simulator up to quickly run them all at once, without any manual manipulation inbetween.