You're appealing to Dawkins, but I think he would disagree with you. It's the teleology again.
Not teleology, I'm still talking about an emergent system. It's just a question of what sort of characteristics end up having a lower chance of ending up as dead ends. Adaptability, including characteristics that improve evolutionary adaptability is a characteristic that helps avoid dead ends. The simple, middle-school version of evolution basically has a model where mutations occur at random. However we already know of systems where organisms have evolved so that at least certain mutations are promoted. For example, spots on animals actually happen through standard mutations that happen at particular ratios. Basically, the organism evolved so that some of its cells always mutate in certain ways. Basically genetic instability and mosaicism, and those characteristics are evolved. That's just a sideshow to what I'm talking about though, and it turns out that the word for what I'm talking about is evolvability. So, things like stress evolved mutagenisis (beyond just epigenetic factors). Some species are prone to chromosomal crossover during gamete combination from sexual reproduction. There's also evolved tolerances to the effects of gene duplication. There's another neat one I just learned about during this discussion, heat shock proteins.
Anyway, you are right that lots of things that are not biological evolution get conflated with it. Of course, that's very tricky because biological evolution is a reaction to environment, and social systems, for example, provide selection pressure, so these systems can co-evolve. A parting thought about Lamarckian "evolution" of course is that, as wrong as he was in general, in some specific ways, his ideas actually do work (for example stress related epigenetic factors that pass on down through generations) in very limited ways. Also, it's very clear that the famous mouse experiment could have actually worked in the end (though it was actually conducted to disprove him). The reason for this is that the experiment went on for 22 generations (though some sources say just 5). The thing that always bothered me about that is that mice have 5 to 8 pups per litter. Even just 5 generations is hundreds of mice, but 22 generations would be what, quadrillions? So, obviously mice would be discarded from every generation, and human bias would be almost impossible. If the bias was based on tail length the tails might actually get longer or shorter depending on the specific bias. In the actual experiment though, apparently the selection protocol was for health, vigor, and fertility. So, even if it didn't affect the tails, it almost certainly would have been selection pressure for other traits. Not to mention that he was almost certainly inbreeding, so that would have had effects as well. Of course, I should note that most criteria for health that would be used in such an experiment would probably increase overall size, which probably would mean that tails would grow proportionately.
Anyway, sorry to ramble, but the potential unintended consequences of those experiments has always been fascinating. Just in general how difficult it is to actually predict every outcome of playing around with emergent processes like evolution/natural selection.
