Presumably the changes to the type signatures in this module warrant a major version bump?

Unfortunately yes.

OK. Can you mention these changes in the CHANGELOG, along with a description of how to migrate code that might be broken by this change? (I'd do so myself, but I'm not sure I understand which situations would break due to these changes.)

If people have Getting [a] values they pass around, they will need to use Getting (Endo [a]), but I doubt anyone does that. Or if they define own combinators (which take in Getting [a] a a), they would need to change signature. So in short: fix type errors GHC tells you to fix.

(It's a unfortunate that Getting r s a has that r argument, leaky implementation detail)

Here is a proposed addition to the CHANGELOG:

5.2 [????.??.??]
----------------
* The type of `universeOf` has changed:

  ```diff
  -universeOf :: Getting       [a]  a a -> a -> [a]
  +universeOf :: Getting (Endo [a]) a a -> a -> [a]
  ```

  In many cases, using `Endo [a]` over `[a]` improves performance. Most call
  sites to `universeOf` will not be affected by this change, although you may
  need to update your code if you define your own combinators in terms of
  `universeOf`.

Does this sound right? If so, I can add this after merging the PR.

that's great! thanks.