Ah, I saw that you did already.

Can you add a note in the issue for that?
Feel free to close this afterwards.

OK, I changed the documentation to state "bound: exclusive bound on plan clost". 
This should probably also be added to the auto-generated wiki documentation 
branch.

Indeed with real-valued costs, which some people have been asking for, it
wouldn't be possible to do the right thing in anytime planners if we made that
change. I think it's best to clarify the documentation rather than changing the
behaviour.

It might be better to change the documentation to say that setting bound=B 
searches for plans of cost < B.
Otherwise, we would need to change the behavior of iterated_search as well (it 
currently passes the cost of the best known solution as the bound for the next 
iteration).
Furthermore, if we ever have non-integer action costs, the current behavior is 
the only correct option for iterated search.

Just my two cents.

Currently nodes with g >= B are not expanded for bound B. Since the bound 
parameter is documented as "bound on plan cost" it would probably make more sense 
to change the test to g > B and allow plans with cost B.

What do you guys think?