To clarify Tom's post just a bit: When he refers to bounding box, that means the area necessary to maintain the Down state, and continue to drag the object. I edited the image, copying the Normal state to the Down state and then sizing the image. When I do that, the bounding area will maintain cursor contact with the image, but only the visible part will register a drop.
The attached sample shows a method of forcing the learner to click on a part of the image that ensures the cursor will be over the shrunken part of the image. It may seem sort of klutzy, but as Tom says, clicking in the area of the shrunken part really is the only way to prevent the separation of the cursor and the image.
On a side note, there is one confusing issue that you can see in the Won't Work slide. If the original click is inside the area of the Down state when it is dropped, it doesn't return to Normal state (sometimes). This whether it is dropped on the target, or drug from the target back to its starting place. However if the original click is outside of the Down state when it is dropped, it returns to Normal state. I can't explain why, except for thinking it is a minor bug.
