1. I think one can make this point, but I also think it's a disappearing problem. I can't state anything too firmly here, but I'm reminded of this interview- http://radar.oreilly.com/2009/07/sequencing-a-genome-a-week.html where the tech guy in charge of Washington University's Genome Center talks about the progress of systematizing genome sequencing and reconstruction. It sure sounds like a lot of progress has been made very recently in terms of systemizing, automating, standardizing, etc, and most of the messy problems have been 'pushed down the chain' into the hands of genomicists.

2. I would say that a first approximation can be very rough, and in this context it could still be useful if one limited discussions to relative comparisons of LED. The other relevant item here is, of course, that the process of trying to simulate genomic evolution could be a very generative endeavor.

3. Both points are very valid. I think validation is a problem with this model, though (as you perhaps allude to re: performance) it may be a problem with other models to some extent as well?

Continuing on (3), I guess my argument would be this: validating accuracy is a problem rather unique to this tool because it attempts to model something empirical, which other tools do not. E.g., existing phylogenetic inference methods assert things, but the meaning of what they assert is more or less definitional. You may view the need for validation/accuracy as a weakness of this tool; I think it could also be viewed as a strength, that it goes beyond other tools in that asserts something that /can/ be said to be empirically accurate or inaccurate.

Thanks for the intelligent comments.

1. Genome assembly is still an art as opposed to a science, especially for the organisms we care about.

2. A successful first approximation to evolution on the genome scale implicitly assumes knowledge and understanding of many processes we do not know and know not of! The concept of selective pressure applied to the genome is an extremely cutting edge field of study.

3. Validation. The problem with any such tool would be validation both on the level of accuracy (how close to the "truth" is the output?) as well as performance (how is it better than existing phylogenetic inference methods?). For the former, the proposed method obviates the convention of validation based on simulated evolution, since such a demonstration would beg the question. As for the latter question, there is no framework within which to posit an answer given the predicament we have with the former.