I agree with your reasoning, I just concluded that it supports the opposite order of things...

The different conclusion is probably because I didn't explain the user interface involved. There are three input fields: One to enter text, one to upload a photo, and one to enter a geotag. All three fields are optional, although nothing will actually be done unless the user submits text and/or a photo, since there wouldn't be anything to geotag in that case.

I figure that, if the user enters anything in the "geotag" field, then that should take precedence over any passively-obtained information and a more-specific entry (#1) should take precedence over a less-specific entry (#2) if both are present. If the user doesn't fill in the "geotag" field, or if their "geotag" entry isn't a recognizable location, then we fall back on information specific to the photo (#3), reserving the user's IP address as an absolute last resort (#4).