Again, all I wanted to know is how confident you were of trusting the bot's rollout. There are many things to pick apart when the bot gives you an answer that you weren't expecting. For this position for example the first things I check are:
- Immediate cube actions after Play A and Play B. Are there any? Should there or shouldn't there be? According to rollouts the proper cube action after either play is ND/T and that agrees with 4 ply so the rollout is fine in that respect.
- Are there any immediate rolls the bot is mishandling after either play? Luckily this is a pretty straightforward position and the bot doesn't have too much it can mess up. However, after making the 5pt the bot on 4 ply seriously misplays 61 and perhaps makes a tiny error with 32. The 32 play is a wash regardless but the 61 play has some influence on the results.
- Can I foresee any errors, usually cube errors, the bot is more likely to make down the road after one play or the other? Here that is not an issue.
I'll check the temperature map in GNU or the dice distribution graphs in XG to further investigate. If I'm really stuck on it I'm likely to perform a stronger rollout than the one that was posted. In this case I decided for shits & giggles to do a longer 3 ply rollout to see where it would end up after the variance was brought down a lot. Posted below, basically the same result.
As Tim mentioned, variants are a good way to tackle some problems. I also love playing out positions.
|
| |||
| XGID=----b-EaDB--cCAa-c-c-b----:0:0:-1:35:0:0:3:0:10 | ||||
 to play 35 | ||||
is Player 1 | 1. | Rollout1 | 18/15 10/5 | eq: -0.747 | |||
| ||||||
| 2. | Rollout1 | 10/5 8/5 | eq: -0.764 (-0.017) | |||
| ||||||
| 1 20736 Games rolled with Variance Reduction. Dice Seed: 21140344 Moves and cube decisions: 3 ply | ||||||
Stick