You are sort of right, but you completely missed the point nonetheless. Evolutionary theory proscribes some large number of small steps between two points. At step(n), selection(n) does not alter the odds of step(n) occurring. In this you are correct. The point you miss is that selection(n) constrains the possible phase space for step(n+1), thereby altering the probabilities for all step(n+k) where k>0 (a recursive feedback loop that reduces the number of possible outcomes at each step, increasing the odds of any one of those outcomes of happening).
This is why the aggregate probabilities are not a multiplicative function of simple combinatorics. You cannot assert the probabilities at each step until the selection function has been applied to the previous step which actually limits the number of possibilities at each step. You have to use the aggregate probabilities of each step post-selection from the previous step which makes each subsequent step far more probable than if you assumed the phase space was unconstrained (which is what you do).
Time for dinner...
I think I get the point of evolutionary pathways, however it seems you missed my point. Evolution theory does not determine how nature behaves. Nature behaves according to its own 'logic'. A theory needs to be in accord with the 'logic' of nature, not the other way around.
I think we are pretty much in agreement that to change a protein into a better one there are very few combinations which will be successful (else you would not need supercomputers to do your work). I also would agree that in a situation where several steps are necessary, the successful outcome of the last phase is constrained by the steps leading to it. Where we disagree is that you seem to think that because the last phase constrains what will be successful, that it pre-determines how the prior steps will take place. Mutations are not constrained by the possible outcome. That is why there are so many bad mutations. While the chances of an outcome being selected may be constrained at each successive change, the number of attempts needed to achieve each step are still determined by the whole phase space of random chances possible in achieving the change. Therefore, increased constraints on what will be a successful change decrease the chances of its being achieved.