This directory contains fake data that should be hard to update without groups.
Each tree has 20 taxa, so tree space is large and T is huge
There are 10 genes, each with only the same three tree topologies and equal weights.
The posterior should be concentrated uniformly on the three GTMs with only one group.
With single gene updates, it is hard to move from one good state to another
as there needs to be a long traversal through low probability space.
But the update groups can jump from one good state to another directly.

Test example data with this command.  Default behavior is good.

  bucky -o test1 ex*/ex*.in

Look at test1.concordance to see that the three uncertain splits each have probability about 1/3.

Compare .concordance results with this command.

  bucky -o test2 -c 4 --no-use-update-groups ex*/ex*.in

The file test2.concordance has the wrong posterior probability, and adding chains does not help.