I am struggling a little bit to understand some of GNU recommendations on doubling.

The one that GNU seems to vary a lot to me and I don't understand is when your opponent doubles and your odds of winning the game are around 33%. At first, I had no clue why sometimes it said to accept and sometimes to reject. I then realized that I was focusing on the odds of winning the game and ignoring the odds of me be gammoned.

So I started to realize that was important. But now I still see variations I don't understand. For example, I have had odds of winning game be 33% and odds of me being gammoned 30%. It said it was a bad play to reject the double. But in another instance, I had odds of winning were 33% and odds of me being gammoned were 20% and it said it was a bad play to accept the double.

So the only difference I can see is the score of the match. Does that effect doubling? I am not talking about when someone is 1 or 2 away. I get that changes things. I am talking about you are playing a 7 point match and your opponent is winning 4-0 versus you are even at 2-2. Does that change the doubling % because you want to be riskier when you are losing like that?

Also, when your opponent is 2 away, GNU seems to feel that the optimum time to double is when your chances of winning are 55%. I was wondering why not 51%, since you are giving your opponent a dead cube. But when your opponent is a 2 away GNU hates it when you double at 51%, but likes it when you double at around 55%, and hates it if you let the winning % get over 57%. Perhaps GNU feels that your opponent will most likely accept the double as long as your odds of winning are under 57% - so don't double at 51% and let it play out a little longer.

Thanks for your thoughts.

Great questions! Looking forward to hearing some of the answers.

Yes and yes.

The relevant term not included in your post is match equity. There are standard tables for this that you can look up.

The key point here is the idea of a "market loser". You don't want to double unless there are some sequences where your opponent will drop if you wait and double next turn. Those sequences are called market losers and they're important. If your opponent could correctly take a double next turn after all or nearly all sequences, then you're better off waiting a turn, letting your position get a little better, then doubling.

If your winning % is only about 51% in the current position, then you probably have very few market losers, so GNU is correctly suggesting you wait on the cube.

Thanks for the information - it is interesting. As a casual player, I find the theory of all this interesting. As a practical matter, I don't think I can apply it.

I read about the match equity tables and the guy gave an example of a match being 3 away and 4 away and how the optimum take point was 23% using the tables, not considering gammons. The calculation was somewhat complex.

Now in an actual match, I would have just done the formula you are taught when you first started playing that the take point is 25%, not considering gammons.

As a practical matter, I don't have the ability to distinguish between whether my chances of winning are 25% or 23%. I have a hard enough time distinguishing between 25%, 33% and 50% and gnu says I am wrong not an insignificant amount of the time even with that. So as a practical matter, these calculations seem pointless in actual play. At least for me.

Craig, we are in the same boat. I find it incredible when I see some of the distinguishing factors between a position that is won, lost or in between, and to recognise those I think takes years of practice.

I think the idea is to realise that that perfect knowledge exists, and through practice, try and narrow the margins between decisions. The thing is, you have to start somewhere... So look at a position, estimate the odds, and get them entirely wrong, but keep doing it. Hopefully you get better at it, and narrow the ranges.

That's my theory - I have to say that casual play and practice fails to deliver any expertise! I commit in fits and flurries, and the results show.

Good luck.