In the context of the IntersectionObserver v2 proposal, there has been discussion about the relationship between an element's layout position with respect to hit testing and its position with respect to drawing. Typically these are expected to be the same, on the basis that users intend to interact with the thing that they see below the mouse cursor.

For IOv2, the hope is that we can detect visual occlusion by hit testing some number of pixels of an element.

Unfortunately CSS and SVG filters break this relationship, because they affect visual appearance but not input events.

Here is an example of a div (the red box) that has a hit test area entirely disjoint from where it appears on the screen: http://jsbin.com/coziqokogi/3/edit?html,css,output

IOv2 can signal if an iframe has a filter applied to it, but we run into trouble if another element elsewhere on the page has a filter that causes it to occlude the iframe. It might be possible to account for filters in IOv2 hit testing, but that would be a divergence from how input currently works.