Like I said, "Sometimes they can prematurely bucket ideas into relevant/not-relevant, but usually not." I wrote that hoping to preempt your exact response. Alas! ;)

Even if I grant your premise (which I don't), just because a breakthrough is likely to be written in crazy-talk doesn't mean something written in crazy-talk is likely to be a breakthrough. At that point it becomes a question of opportunity cost, which is exactly where heuristics come in.

In fact, what you said could be true and it could still be true that

  P(breakthrough | does not seem crazy) > P(breakthrough | seems crazy)

So, if I were on the hunt for breakthroughs I'd still be better off ignoring the crazy-seeming things.

Regarding your premise, can you name, say, ten mathematical breakthroughs since the Enlightenment that came about the way you described?