Dr Zare has completed a thorough analysis of 22 matches played by
Hank Youngerman (RedTop) and concluded, "I'm over 99% convinced
that Hank Youngerman cheated based on the match records."
On February 27, 2006, I (Paul Weaver) sent an e-mail to Hank
informing him that I knew he was cheating at backgammon
online. I had no idea it would ever be publicly posted. I sent
copies of the e-mail to about half a dozen friends, and a few
days later it was posted on this forum (not by me).
A few people defended Hank. If there is anyone left who still
thinks Hank is legitmate, he should read the analysis by Dr
Doug Zare that begins here:
"After reviewing some of Hank Youngerman's matches, I have concluded
that he was cheating systematically (as opposed to cheating
occasionally on tough decisions), probably using Jellyfish. I base
this conclusion on the following pieces of evidence:
1) Youngerman had an extremely low *checker play* error rate when
measured by Gnu Backgammon.
In 22 matches, his average checker play error rate according to gnu
evaluations was 1.28. For comparison, the best checker play error
rates according to matches sent to Karsten Nielsen are those by
Francois Tardieu (2.25) and Falafel (2.39):
http://www.gammonluck.com/ranking.html. Yours is 2.60. Wachtel's
checker play error rate over 17 matches against Youngerman was 2.44,
though his error rate on Nielsen's list is 2.90. According to the
calculations I did for my June 2006 column, my checker play error rate
Youngerman's overall error rate looks less suspicious since his cube
error rate is relatively large. Checker play errors are more
important, are more consistent between matches, and are a more
reliable indicator if someone is playing in a bot-like style.
If someone runs a 6 minute mile, it is a sign that he is in good
shape. If someone runs a 4 minute mile, he either cheated or he is a
world-class runner in peak condition. If someone reaches the finish
line in 2:30, look for a car.
This sustained checker play error rate of 1.28 is like running a mile
in under 2:30. It is much better than anything sustained by top
players. It is not explainable by the sample size, match length, or
that people tend to play better against strong opponents.
2) Yougerman's checker play error rate is even lower when measured by
His error rate over 22 matches was 0.59. This is a cubeless checker
play error rate, and is not directly comparable with the cubeful
checker play error rate measured by gnu. However, a reasonable
expansion factor is about 1.4, which would make the comparable figure
Youngerman plays significantly more like Jellyfish than like gnu or
Snowie. To a small extent, this can be explained by his use of
Jellyfish as a training tool. However, an error rate of 0.83 is
inhumanly low, less than 40% of the analogous error rates of the best
3) Youngerman makes few substantial errors according to Jellyfish.
In the 22 matches analyzed, Youngerman made 0 checker play errors of
size 0.050+ according to Jellyfish, and only a couple of errors
between 0.040 and 0.049. This is strong evidence not only that
Youngerman got help at least some of the time, but that he had help on
the vast majority of the moves. Even top human players often make
large blunders due to overlooking plays or misjudging the strengths of
positions that are hard to compare. The analogous cubeful size is
about 0.070; top players (including Tardieu) routinely make blunders
larger than 0.110 every few hundred moves even when they have low
error rates. It is extraordinary that Youngerman made no checker play
blunders, and nothing close to a blunder.
Jellyfish has a mode in which it will alert you only if you are making
a mistake larger than a particular threshold
(View->Settings->Comments; uncheck Show best moves, check Comment
moves, choose threshold). I believe Youngerman either used this mode
with an error threshold of 0.020-0.040, or else intentionally
introduced what he believed to be small errors based on Jellyfish's
4) Youngerman made many strange bot-like plays.
Most human plays can be justified by their accomplishments. Even the
errors can be explained by paying too much for some assets, or
overlooking the game plan. However, some of Youngerman's plays are
hard to justify ... except that they are Jellyfish's top choice. There
was often little equity at stake, as there were other (more natural)
plays which were as good.
5) Many of Youngerman's take/pass errors agree with Jellyfish.
Jellyfish does not understand match play well. According to more
modern bots, Jellyfish takes too deeply at 3-away 3-away, for example.
In many of the positions where Youngerman was flagged by gnu for a bad
take, few human experts would take, but Jellyfish would have taken,
In conclusion, I'm over 99% convinced that Hank Youngerman cheated
based on the match records.