This is a very simple problem in all honestly (I should hope it is as I am currently doing a degree in it), I can see jck that what concerns you is the chance of extreme mutations occurring , and I understand it does seem extraordinarily bizarre. However, using your fish as an example, before the first stage of mutation, all the fish were capable of surviving however, due to their lack of camouflage maybe 10% were killed every year due to predation (purely hypothetical of course) lets say then, that a mating pair spawn (via mutations during gametogenesis) an individual with slightly differently coloured fins (this could easily occur, most likely as one or two amino-acids on a single DNA stand of a chromosome being altered), this would potentially increase the survival chance of the fish, by lets say 1%. out of a hypothetical population of 10,000 there is to be an expected death rate of 1,000. In all likelihood our individual fish with his SLIGHTLY improved genetic adaptation is very unlikely to die (if he did, then of course, the process would end until another fish with similar/same mutations occurred).
Our fish being better adapted to survive now, lives until mating age, at which point from the offspring, some would carry the same genetic trait (depending on whether it was a dominant or recessive trait then the offspring rate would be either 50% or 0%, however the 0% would all be "carriers" of the genetic sequence, which could then be passed down and possibly be later shown in the phenotypes of later offspring) will assume that this is a dominant autosomal gene, and therefore 50% of offspring now carry the trait, due to the large offspring figures we shall say that 100 offspring are spawned. obviously not all will survive to full birth (i.e. hatching) so we shall say only 50 survive to true birth.
we now have 26 (including parent) mutated individuals all with slightly elevated chances of survival. Over hundreds of years obviously these numbers will greatly expand, and eventually another mutation may occur, that lets say made the tips of the fins slightly more rounded, or to have a slightly differing pattern, there will be other mutations that are completely useless, however these born individuals, have similar or lower chances of survival compared to their ancestors, and are therefore less likely to survive.
Hopefully you can now see that it actually makes perfect mathematical sense, I would advise reading some literature on "Mendelian genetics", which shows the basic mathematical principles. which is amazing considering that Mendel was a monk who came up with his ideas before microscopy was even invented, and his ideas have shown true even today!
P.S. sorry if this is a bit of long winded response, but then genetics ain't a simple topic!