Please, reread what have you written before. How can one "test within a very
good degree of likelihood whether or not the command will
execute properly " without knowing the connection state?
Not to defend a self-admitted troll, but...by having known common
failures.
If a particular condition always causes a failure, and that condition
is the cause of X% of failures, you can avoid making an attempt that is
doomed to failure by first testing the condition.
If X is the vast majority of failures, then your confidence in the
success of an attempt can reasonably be tied to whether or not the
condition exist.
So, empirical data can tell you where to put your effort.
(On the other hand, I think your basic point was that exceptions
shouldn't be used when there is a more straightforward way of returning
the same information is valid).