I read this SRFI with interest and here are my very minor comments.

  - Ref. "Draft #2 published: 2018/11/11"

  - (FYI) I tested the sample implementation in Kawa and all tests
    succeed.

  - (Question) The document notes, "It is an error if two threads
    access the same equivalence class of unifiable boxes at the same
    time."  Does (or should) that mean it is guaranteed not an error
    for two threads to access distinct equivalence classes
    concurrently?

  - (minor) It may be helpful to include the keywords "union-find" and
    "disjoint-set" in the abstract so that someone looking for a
    union-find data structure will be led to this SRFI.  The Rationale
    mentions it right away, of course, but some (tools/people) may
    only search abstracts.

  - (minor) It may be helpful to explicitly mention that the return
    values of ubox-set!, ubox-unify!, ubox-union!, and ubox-link! are
    unspecified (perhaps just once for all of them).

Regards,

-chaw