So that's the question. Sometimes Gnu-backgammon will criticize my cube action. But sometimes I am trying to play correctly. I am trying to learn.
But other times I am just trying to win. And the computer is a lot better than I am. Just like in a tournament I know some people are better than me. So I may make an aggressive speculative cube. The computer thinks that is a horrible error.

Is it correct, or even really possible, to adjust your play (both cube and checker play) when you are the weaker player? If you really consider that question it really requires a fairly high level of skill to adjust your cube action. Anyway the basic question holds of whether being the weaker player affects cube play

The only thing I would say the cube is the most difficult part of the game. Also, what wrecks your PR too. It's a mystery when just 2 pips is good enough and when 20 just not. So we are in the same.
So what Im doing is taking any cube decision I see in my games or around, and making a reference cube collection. It would be great if there were some kind of online site like lichess.org that allows publishing your studies and analysis and everyone benefits from it.

There's no reason to play that way vs a computer. Computers are for practice and learning to play correctly. In a real match if you are confident that you're the weaker player then yes ramping up variance makes some sense

I generally agree. This only came up because I wanted to end a match early win or lose. And of course the engine is always going to mark dubious plays as such.
However the analysis made me consider whether such ideas ever really make sense. That is even against human players. I think you'd have to be quite a strong player to properly adjust your cube decisions to comparative playing strength. Most likely it loses more equity than it gains. I was just wondering what others thought about it.

I don't know if I'm understanding this discussion correctly. But if I am a weaker player than the bot, then there would be close situations where a bot should double and I should not. The reason is that the bot is making a decision based on how likely it is to win from that position. It is not looking at how likely I am to win from that position, and being that I am a weaker player, it might not be best for me to risk playing a doubled game. . Well, unless is is factoring in my playing history I have with the bot, and I don't know if they apply that to doubles.

I think we are discussing match play here mainly (and the idea is simpler to understand vs money play). Assume you are playing a 3 point match against a top-level pro or a bot. In one match we can play normal backgammon rules; cube in play, gammons etc. in another.there are no cubes or gammons; each game counts as 1 point. Which match gives us a better chance to win?

Well, it should be obvious but the normal rules give us a much better chance (especially against a bot who will use optimal cube strategy and not adjust for skill difference). Say you are bad enough that you will only win 30% of the time on average. If you must win 3 out of 5 games, you have about a 16% chance of winning. Under normal rules, though, you can double at first opportunity and win 2 points when you win. This effectively changes things from needing to win 3 out of 5 to a match where you need to win 2 out of 3. You would have a 21.6% chance in that case.

You could do even better with a more optimal cube strategy, and in this example you could almost guarantee at least a 25% chance of winning almost all the time. If you play a bot, it will double when it has a 75% chance of winning the current game (give or take a bit). If you get doubled you could simply redouble immediately and play for the match with your 25% chance of winning.

In general the principle holds that in any unfavorable gamble, you have a better chance of coming out ahead by increasing the variance. You will never beat a top pro poker player, for example, by playing 100K hands using a normal strategy— the variance gets washed out over that sample size. You can give yourself a pretty decent chance to beat him in a heads up tournament type game by playing an all in pre flop or fold strategy. At worst, even if he knows what you are doing and only calls with premium hands you can give yourself at least a 20% chance of winning — the variance is high.

Getting the cube to high levels increases variance. You want to increase variance. Your best chance of winning is not to play safer, but to play riskier, especially if you do manage to put yourself in a favorable position

Well, yes, all of that. My point was actually very simple, so I'll add more context to remove more variables. A 21 point match with a score of 3 to 3. XTG says that if it were in my position it would double based on its winning chances. I might not double because my winning chances are actually lower. That's it. And so barring other variables that might enter into this, I might not double when XTG would. Of course, for me to learn I would sometimes follow it's advice to see what follows. But that's another story. I'm talking about just playing a match.

As the weaker player, yes you will win less often in such a spot than what the program thinks you will. You still should double. Even winning less often than optimal, you are winning in this position more often than you would be in subsequent games in the match. Look at it this way: shorter matches give the weaker player a better chance to beat the stronger player. Doubling in such positions has the same effect. The more games you need to win, the lower your chances are. By doubling you reduce the number of games you must win.

Additionally (this likely is not a critical factor early in a long match) the program calculated doubling points assuming each player has a 50/50 chance in subsequent games. If this is not true, then the doubling points will be wrong, maybe significantly so. Generally this also trends toward the weaker player doubling lesser positions as well.