A couple of things:
1) Mohan, for Dretske's view on inattentional blindness, his paper in Phil. Perspectives 2007 is useful. As for the Zorro case, I meant "distinguish" in a non-perceptual sense (perhaps wrongly assuming that Benj's notion wasn't particularly perceptual). For a similar perceptual case where, S visually differentiates x from y without knowing they're distinct, one might imagine that S has a hallucination of the pink background surrounding the kangaroo that's really in front of her (or alternatively, a giant bright yellow truck and a giant blue truck in the periphery of S's visual field, when S is fully focusing on and thinking about the kangaroo she is foveating).

2) Hugh, indeed. I thought differentiation was meant to be a necesary condition on seeing (though the formulation Dretske uses "=" is a bit unfortunate), and the first clarification he makes about the notion is that it'd be visual (check out p.20).

3) Benj. That's the limiting case I mentioned earlier. He discusses this sort of (no background) case on pp.24-6: upshot is, S sees(ne) x only if, unless everything in front of S is part of x&conditions are such that x has no environment, S differentiates x from its immediate surrounding. (Not saying this works, only that such cases are taken into account.)

4) Benj, on distinguishing and knowing distinctness. Why do you need the right-to-left entailment? Doesn't seem to play a role in your objection to the differentiation constraint + possible sense of distinguishing where: Matt knows the twins Aaron and Bryce are distinct but he simply cannot distinguish them when he sees them.

5) Benj, about looks: not sure there's any disagreement, but not sure I get all the bits in your proposal (relativized judgement? Is looking at o when o looks F sufficient for judging that o is F? Is looking at o when o looks F sufficient for selectively attending o being F and for noticing that o is F?).

I thought the  following was rather trivial. If C = whatever conditions are sufficient for S to be caused to believe that o is F (or looks F) on basis of S's experience of o when o looks F to S , then we can analyse looks in doxastic/gnosic terms: x looks F to S --> if C, then S believes o is (or looks) F. But, I assume, that's not what you take yourself to be doing, or is it?

+ on formal framework: I'm sure there's different ways to do this (long conditional, centered worlds, other contextualist approaches). One thing I don't get without hearing more is: is there any non-independent (e.g. absolute love of centered worlds in general) reason for preferring one way to do things over another? That is, any reason against conditional approach in this particular case?

Finally, not sure if this account of looks helps in the objection that the differentiation constraint is implausible (not sure whether it was meant to): seems compatible with S not actually believing anything about the gorilla and its background that S might.