Maybe lossy is the wrong word, but the transforms historically had gotchas. PPX and what not. Especially when converting between Reason syntax version upgrades and the more stable OCaml syntax (locked to OCaml 4.2 because of bucklescript).
That doesn't necessarily mean lossy, but just that knowing what refmt will error out on depending on the version of Reason the code was written to target vs. locked down ML.
the explicit_arity being added in. I can't think of other examples at the moment.