It's possible to make observations about other observations (e.g. assertions of clinical significance/importance and other interpretations that happen at a time that's separate from when the original obsrevation was made). It's also possible to make observations about practitioners, organizations and all sorts of other things. This has been done in v2 (with OBXs accompanying all sorts of things in v2 message structures) and it will be needed in the FHIR space too. There will need to be guidance on when to use observations vs. extensions. The guidance is that observations stand alone and capture information about the provenance of the assertion (who made it, when, how, etc.).