[whacked_new]

You asked why the learning is not taken as an aversion response to lack of reward. My interpretation is because your baseline (square) is zero: that in itself is not aversive; it's just nothing.

I think the new rectangle is evidence of hyperstimulation, where a stronger response is elicited when features are exaggerated. To establish this case, we'd have to test how strongly the rat perceives prototypes of squares vs. rectangles; my guess is square-ness vs. rectangle-ness is easily discernable. That is to say, I am pretty convinced this is just hyperstimulation, in which case a random shape would not do. It would have to resemble a former stimulus X but exhibit features that are even more X-like than the previous Xs. If perception of a rectangle is encoded as "right-angled shapes where height != width" (a plausible encoding, because our shape detectors operate on angles and lines, while curves utilize different detectors), then when height !!!=== width, then you get BAM!

An analogous example in humans? Tomb Raider :-)