Issue 1

Physical screen pixels may be ineffective. For example, the physical size of the contact cannot be surmised due to varying pixel densities of devices (which are not always reported by the user agent). Further, it is generally impossible or difficult to correlate this back to CSS pixels in the document (which often are not at a fixed ratio to screen pixels due to zoom).

I think the starting point should be that you can use the API to infer whether a contact overlaps a particular element's bounding box. With device pixels, this is difficult or impossible.