I found this quite an interesting example of a genetic algorithm. I had so far just seen them in theory, and the one time I tried to implement performance was too slow to be practical. I had never thought of implementing this in T-Sql, but it shows you can do a lot with it.

I was wondering 2 things though:

- wouldn't a C# implementation be much faster? (if I can spare a few hours I might actually try)

- I may have missed something, but I didn't quite understand what the use is of random moves. Wouldn't that influence the rest of the route, and more importantly make the fitness function non-deterministic?