Mochy's backgammon study: Last problem of "100 backgammon puzzles" by Paul Lamford

is Player 2

score: 0
pip: 60

Money session
Jacoby Beaver

pip: 330
score: 0

is Player 1

XGID=-------------------eEeEeE-:0:0:-1:D:0:0:3:0:10

double to 2 take ?

Analyzed in Rollout
No Double
Player Winning Chances:	83.98% (G: 56.17% B: 30.61%)
Opponent Winning Chances:	16.02% (G: 0.01% B: 0.00%)
Double/Take
Player Winning Chances:	81.40% (G: 69.01% B: 39.53%)
Opponent Winning Chances:	18.60% (G: 0.14% B: 0.00%)

Cubeless Equities
No Double:	+0.680
Double:	+3.424

Cubeful Equities
No Double:	+1.024
Double/Take:	+3.257 (+2.233)
Double/Drop:	+1.000 (-0.024)

Best Cube action: Too Good to Double / Drop

Rollout details
1296 Games rolled with Variance Reduction. Dice Seed: 1341 Moves and cube decisions: 3 ply

Confidence No Double:	± 0.017 (+1.007...+1.041)
Confidence Double:	± 0.136 (+3.121...+3.392)

Double Decision confidence:	99.7%
Take Decision confidence:	100.0%

Duration: 20 minutes 42 seconds

eXtreme Gammon Version: 1.21

This is a last problem of "100 backgammon puzzles" written by Paul Lamford. Paul says it is very easy take there fore no-doulbe is proper cube action. XG rollout (3ply full) is strongly against it.
However, I coudn't trust bot's recommendation of "too good to double" with Jacoby rule, centerd cube. Who is wrong, bot or Paul?
I set up XG playing each other to see how bot plays. I understand why it was too good. 3ply bot will take the bad cube later, so that doubler can gain by waiting one more roll.
Also, 3ply can't handle the outside prime("snake") properly. As a result, take side simply wasting TOO MUCH equity in the process.
The position below is ending scene of rollout and good sample of how bad bot plays. Human wouldn't stack 7 checkers on the 18point in this game.

is eXtremeGammon

score: 0
pip: 10