Isn’t canonicalization not good enough? Even if you canonicalize a spider trap, a robot can’t (or won’t — if you’re not important enough) spider a spider trap.

Implicit selection is also only going to happen if it’s a single selection interface. Along those lines, it’s probably best just to exclude multiple selection entirely from the view of search engines. It’s just too huge a space.

Canonicalizing a spider trap is like saying “see all this stuff you spidered? Yeah, it’s useless.” Isn’t it? I’m assuming Endeca sorts the filters when they beautify URLs.

Canonicalizing solves small duplicate content problems — like deduping a product in N categories. Facets introduce powersets and factorials, and that’s a bigger challenge to throw at it. Perhaps too big, no?