Extreme self-commenting code.
Not going to read all this but:
Naivety. Sure self-commenting code helps a lot, but the very best most efficient code can be extremely complex to read and understand (so comment where necessary). The smarty pants that wrote it, probably factored in one step what would take others several steps to do, so how do you expect others to suddenly get the insight he has in factoring to understand the self-commenting code. To further add to the problem, said smarty pants, if he did comment it, might think it is so obvious that his comments are so abstract that they don't help much either. So, comment where necessary down to the standard the competent/average programmer can understand (which ever is less). If you have below average programmers, comment down to their standard. And don't assume that a competent, or average, programmer bears an direct relation to your high standard of intellect.
I tend to write down to a level of reasonable readability. But I had an engineering friend in uni, and she could immediately write something simpler than several of my attempts. Even in my VOS design documents, I sometimes write descriptions that seem to have obvious meaning, that I cannot fathom latter. The problem is we assume something is obvious because we already know the knowledge experience or answer, so we should never con ourselves that the meaning is so obvious.
One of the problem I imagine, in this debate, is the common use of high level language. Where convoluted code can be inefficient code, and the assumption that truly efficient code can be written in that language, leading them to the false security that self-commenting code is more effective. Low level languages however, tend to allow a broader range of possible combinations of instructions and techniques to produce more efficient convoluted code.
I know somebody that wrote an IDE hard disk interface driver in 100 instructions, or 100 bytes, I forget which. The code was not as readable to others as it might have appeared. Actually, the self-commenting code of this class of extreme programming is not so readable.
It is like an ego thing, "yes, spend ten years of zen to understand my program.." The last rule is, why put people through all that pain to understand your code, when you can just adequately comment it for them to zoom in on the area of interest and change it. "Because I don't like commenting.." maybe, in other-words couldn't be bothered, it is somebody else's problem.
When they can extreme pogrom in assembler so that 90% of programmers can easily follow what they are doing, then I'll believe some of these things.
PS Jame's lead looks promising.