The problem with the 'fish-to-amphibian' example is, Tuppence, that you have already established what the end result will be.
You reverse the order in which evolution works, which makes leads to calculations that seem highly unlikely.
It's like the example I gave before, about flipping a coin 10 times. I'll get 10 results which are either "heads" or "tails". REPRODUCING this list exactly by flipping a coin of your own is almost IMPOSSIBLE, but I just did it right the first time! Do you see my point? Evolution is not just 10 flips, it's millions of flips. And the possible results are not 'head or tails' but a huge variaty of possibilities, depending on the animal.
Calculating what the chances are for a fish to become an amphibian are therefore bound to lead to 'unbelievable' odds, just as my row of heads and tails.
I could have thrown thousands of other combinations of heads and tails, and fish had billions of other shapes and forms they could have evolved into. Calculating the odds of one particular path makes no sense.