I have often thought that Snowie's cube decisions from rollouts were
suspect, and did not properly reflect the results of the rollouts. I
couldn't place my finger on exactly what was wrong. This last weekend, Neil
demonstrated to me just what the problem is, with the following example:
Money game, cube in center. X on roll has 4 checkers on ace point and 1
checker on 5 point. O has 5 checkers on ace point.
13 14 15 16 17 18 19 20 21 22 23 24
++++++++++++++
   O 
   O 
   O 
   O 
   O  ++
     1 
    ++
   X 
   X 
   X 
   X X 
++++++++++++++
12 11 10 9 8 7 6 5 4 3 2 1
X doubles. Should O take?
It is clear that O's recube vig is tiny. The only way O could get a recube
in is if X rolls two consecutive 21's and O doesn't roll a double. If this
happens X can cash and not have to sweat O rolling doubles on the last
roll. On any other variation X will take at least 3 checkers off in two
rolls, so either O rolls a gamewinning doubles or he loses but he doesn't
get to recube.
According to Snowie, X will win this game played to conclusion 77.6% of the
time. I believe this is an accurate figure. Therefore, O has a clear pass,
since O wins substantially less than 25% of the time and his recube vig is
virtually nonexistent.
However, Snowie evaluates this as doubletake!
What is wrong? According to Neil, the programmers do not attempt to
determine the likely efficiency of a redouble when evaluating a pass/take
decision. Instead they assume an average efficiency for all positions,
which results in the breakeven point (assuming gammons aren't in the
picture) of about 21.5%. Since O wins greater than 21.5%, Snowie thinks it
is a take.
This flaw undoubtedly leads to very inaccurate cube evaluations by Snowie.
If the position is such that an efficient recube is very unlikely to occur,
Snowie will be taking too often. On the other hand, if the position is such
that an efficient recube is more likely than normal, Snowie will be passing
too often.
I have never respected Snowie's cube evaluations even after a rollout.
Instead, I look at that wins, losses, gammons, and backgammons. These I can
trust, assuming that Snowie can play the position decently which is almost
always the case. I then use these numbers to form my own cube assessment.
Kit
