The models never give guarantees, and it isn't reasonable to expect
otherwise, because guarantees are from the realm of deterministic
results which reside outside the universe of probabilities. Perhaps
the key is to avoid thinking that "predict" and "predict with absolute
certainty" are the same concept.
And that has been my argument throughout this, although we still are miles
apart as far as short term expectations are concerned.
To use an analogy, the models are a sort of map. If we drive from L.A. to
Vegas we know with absolute certainty what we will see along the way and
that given enough time we will arrive at our destination. But the problem
is that along the way, on each trip we make, everything we expect to see is
in a different location! Whoa, Barstow isn't where it was before...this or
that hill has moved...etc etc. So while you know the destination is fixed
and the things you see along the way are always there, you have very
little information about where those things have moved to along the way.
Thus, short term expectations are virtually nonexistent, while the long
term expectation is constant.
And to answer a point you also raised, of course it is important to play
optimally for one reason alone: playing optimally will generally keep you
in the game longer so that you'll be there for those fortunate quirks of
variance. That, in itself, is reason enough to adopt what is often called
perfect play. Expectations of seeing something approaching the normal long
term distribution are not a significant factor in that decision. You WANT
the deviations.
And just a note on the EV/ER thing. Again, I purposely avoided using EV to
avoid bringing other factors that might have the implication of a player
spending his or her way to a goal, and so keep it completely within the
realm of probability, exclusive of any other factors.
lb
"variance giveth, variance taketh away"