In my last experiment, I had added extra challenge for myfelf by
having to play my first rolls forcedly from my 24-point.

After that, I wondered what would happen if I played forced moves
from 13, 8 and 6-points using up both numbers as much as possible
if legal moves allow.

From 13 and 8 most rolls played well enough but from 6 most rolls
caused huge blunders and equity losses. So I opted to go with that.

If I open with 31, I have to play 6/5 6/3 or 6/2, not 8/5 6/5. If
I open with 33 or 44, I have to play 6/3x4 and 6/2x4. If bot opens
and unblocks my 1-point, then if I roll 55, I have to play 6/1x4...

Ouch! Ouch! Ouch! Very few rolls play well. Most play very bad or
worst possible. It may be difficult to imagine without looking at
some actually played games or trying a few games yourselves. The
effect was so bad that XG cubed before the forth roll in 13 games,
while I could do the same only once.

For almost six weeks, I played hundreds of games to figure out how
to recover from opening anti-jokers, to not lose too badly and to
win if possible. Even though this experiment was harder and slower
to play, I stil was able to finish it within two days.

My errors rates were 18.56 checker, 25.21 cube and 19.55 overall.
Compared to previous experiment's 13.89 checker, 31.09 cube and
16.91 overall. Higher checker error as expected and lower cube
error because I didn't get to cube much (and XG didn't get to
beaver as often, so total points won was lower for both sides).

My effective net points won was 126-119=7 vs. expected -114 (i.e.
+121 more points won than I should have according to the bot). At
highest I was 39 points and lowest 2 points ahead, never falling
under zero or behind the bot.

Very similar to my previous sessions, I had 0.0062 average luck.
I won 34 games vs losing 66, of which I lost 8 while I had zero or
positive luck but I never won a game with zero or negative luck.

Interestingly, I won the first roll 44 times vs the bot's 56 (14
of which were in a row of a single streak!)