I broke SNG wizard :D

Originally posted by meepwn
Goes to show that the basic algorithm of the program is flawed and should not be trusted.

Interesting post all in all. But how did you get to this conclusion? *boy, that escalated quickly*

It's not that bad advice IMO. But there is definitely a problem with the algorithm. SNGWiz (et al) compile stats/hand histories for chip edge to figure out this situation, since there is no good way to calculate this.

But here we are with gambling around 0.01% of the prize pool, so an edge of 0.25% makes no sense, despite it's applicability to most big stacks in this situation.

So is it broken, yeah.. but AFAIK there is no better way to calculate chip edge other than statistical situation matching, and that's better than nothing, since Nash ranges don't penalize/advise gambles at any particular time because there is no chip edges.

I think the basic mechanic of getting an "edge%" number, getting a "diff%" number and adding them is wrong. I mean, this is a pretty extreme situation, however all the suggested variables are OK and we STILL get obviously wrong result.

Just a few thoughts:

- mayyyybe the "diff%" should be calculated for effective stacks? Then again, it would be -EV to call a short stack push with AK if we actually have a vulnerable middle-sized stack

- maybe the edge% should be calculated in a different way? Having absolute, and not relative values seems to screw up the calculations

....There is no mathematically sound model of SNG's, and whoever makes one first is gonna make a lot of money.

This is a pretty edge situation. It's not often you have a huge lead like this when the bubble factor is so high (overall). I can see an edge% of 0.25% being good advice to most big (but not retarded big like this) stacks for bubble play of a DON. It's rare that calling or folding to a shove only risks 0.01% or less of the prize pool (in an STT).

- mayyyybe the "diff%" should be calculated for effective stacks? Then again, it would be -EV to call a short stack push with AK if we actually have a vulnerable middle-sized stack

Thats not how I understand diff%. It's not something that you can change its just the difference in EV% of pushing/folding GIVEN we have a particular hand. Then if that particular hand's diff% is above the edge% it's a push.

The edge% is really the only thing we have control over. The diff% is effectively the nash strategy EV for the particular hand. So the question here is why is the edge% recommendation so bad and how can we fix it..?

-edge%, defines our edge over the opponents. If we are playing against good players or if we are UTG with small stack its negative, if we are playing against fish or if we have a big chip stack its positive.

I see edge% as our risk aversion, the lower the edge% the more willing we are to gamble, and if it's high we don't need to gamble as desperately.

The whole point of edge% IMO, is to approximate full depth future game simulation. There exists an edge% such that the resulting strategy is the same as a full depth future game search (ignoring general errors that all ICM calculators make).

I think they give these chip edges really naively, say, if you are over 30bb in a DON, then your edge% is 0.25%, with possible some ICM related changes. But I don't think it's that sophisticated. Most people in this spot will be gambling with around 2% of the prize pool, and 0.25% edge makes sense, but clearly when you gamble with only 0.01% an edge of 0.25% means you can never call, because even if you have 100% equity you aren't making at least 0.25% here, but only 0.01%.

One thing they can do is NORMALIZE edge%, It doesn't really make sense to have only one edge%. Clearly if you are gambling with small amounts, you don't need that big of an edge. But if you gamble with larger amounts you need a larger edge. This would be a much more robust model (even for regular situations), but of course it much more difficult to implement and understand (compared to edge%).

Moral:
Always look at edge% and make sure it makes sense in whatever spot