Evolutionists and design theorists often have an extremely different notion of what constitutes "proof" of evolution, and this causes a good deal of confusion. Evolutionists complain that the standard of proof required by design theorists is unreasonably high, and design theorists complain that the proofs of evolution in the literature are not proofs at all. You can guess where I fall on that spectrum.

If you pay close attention, whenever evolutionists are engaged in listing the "proofs" for evolution, they almost exclusively list evidence for common descent. Ask them for the mountain of evidence that necessity and stochastic processes are responsible for making the transitions between the similar forms which they just showed you, and they may draw a few pictures and connect them with arrows, but that is all you are likely to get.

Take this model of the evolutionary origin of the bacterial flagellum. It is long, detailed, and testable. But just what aspects of the model are testable? Precisely the aspects related to homology. The testable claims are about which components may be more ancient that others, which structures may have been intermediate between the Type-III secretory system and the bacterial flagellum, etc. The question of whether natural selection acting on random variation is powerful enough to make the hypothetical transitions is never addressed. Instead it is assumed that proposing intermediates and telling a story about how each might provide an incremental fitness benefit is enough.

I'm not convinced. You see, I'm a computer scientist. I'm used to looking at incredibly complex, interdependent systems at a high level of abstraction and implementing modifications at the very lowest level. There is very rarely a correlation, even in the most well-designed systems, between what changes appear straightforward at a high level of abstraction and what changes are actually straightforward to implement (by "straightforward" I mean changes which do not require a large number of compensatory changes to other parts of the system). There is not enough detail in the above model (or in most evolutionary models) to evaluate whether or not natural selection is sufficiently powerful to get the job done.

This raises an interesting question: How specific does a model have to be to be convincing? That is, how small do the proposed changes have to be to render it likely that natural selection could have filled in the gaps? I don't know of any formal answer to this, but whatever the answer might be it must be able to test natural selection, not merely assert its ability to move between hypothetical intermediate forms described at a high level of abstraction. Ironically, the only people who seem to be interested in doing this sort of thing are design theorists.