I spent a bit of time digging into aeson's use of dlist this morning (I didn't add it myself, hence the need to dig), and it actually uses almost the entire API. I do not plan to duplicate dlist into aeson, so the remaining options are to bring dlist into the platform (which is fine by me), or omit aeson. The dlist package is somewhat widely used, and it's very stable, so I am unfussed about its inclusion.

Regarding scientific, it solves a real problem with correctness in previous versions of aeson. It is admittedly both new and nichey, but I would be fine to see it go into the platform for the same reason we were okay including the primitive package. It is depended upon by both the latest or HEAD versions of attoparsec and aeson, so the option to fold its code into aeson doesn't really exist.