Worse, one needs to have a probability estimate for every possible route in genome space between ancestor and progeny. A given molecule may have occupied more than one completely different function between its role in ancestor and its role in progeny. Most of the homology data we have suggests that molecules were often coopted; that a molecule, say, that functioned as an ion pump may have fortuitously added another domain and became an ion-gradient-driven mechanical device. It's improbable that a complex multimeric protein came together all at once to fulfill a role; it's much less improbable that a simpler molecule evolved to fill some role, and that a mutation caused another domain to stick, which gave it second (probably inefficient) role, which then evolved. We're a hundred years away from being able to map out the probabilities of such events.
On the other hand, I'm dubious about the claims made on behalf of the neo-Darwinian synthesis, partly on the basis the fact that I hold a probably extreme view of falsifiablity and the predictive part--natural selection--usually ends up being tautolgous, and partly because the other part 'random variation' either has randomness defined away in any statistically meaningful sense, or is untested.
Popper began with a similar view, but changed his mind.
IMO evolution makes plenty of predictions; we can make concrete predictions, for example, about the likely characteristics of as-yet-unsequenced genomes. Using maximum parsimony, for example, I can predict the likely sequence of any even moderately conserved gene in the last common ancestor of humans and rats. You can argue that that long extinct animal will never actually be sequenced to test my prediction - and that may be true (though we are getting better at getting DNA from fossilized material) but we can look at other descendants, and see if their genomes are consistent with that of my predicted ancestor.
It would have been a much less accessible movie than Jurassic Park, but if you have enough different descendants, you could plausibly propose to reconstruct a major part of the genome of a long extinct common ancestor, just by looking at the sequences of the descendants. No mosquitoes required!I can tell you, for example, with a high probability of correctness, the chemical structure of the myoglobin of the last common ancestor of the whales, even though no living whale has exactly that myoglobin. This sort of research is in its infancy, but from the properties of such gene products, we may be able to tell, say, if the ancestor lived on the land or water, how big it was, etc.. That, IMO, is quintessentially predictive.
This is part of what drives me mad about 'the debate'. Both sides consistenly obscure the distinction between the three meanings of 'evolution' I mentioned in my semi-defense of Behe. The first (allele and phenotype dynamics) is simply an observable fact; the second (common descent) is plainly a solid falsifiable (a single organism with a different system of codons would suffice) scientific theory and every observation to date supports it; the last (neo-Darwinism) is the one I'm not even sure manages to be a theory (though once one vacates 'random variation' the way the definition at http://evonet.sdsc.edu/evoscisociety/what_is_evolution.htm does) and leaves natural selection as a tautology (rather than 'so formulat[ing it] as to be far from tautological') which seems to be the tendancy, lest a falsification of a particular such formulation give aid and comfort to religious obscurantists, it's a 'fact' too (but not a very interesting one), provided one drops the insistence on its completeness and sufficiency as an explanation.