Elixir, LFE and a lot of other languages past and present that share the BEAM use this approach. As far as I understand it, the BEAM byte code isn't documented

Elixir, LFE and a lot of other languages past and present that share the BEAM use this approach. As far as I understand it, the BEAM byte code isn't documented all that well and emitting Erlang terms is more straight forward. This approach also makes keeping your language up to date with the BEAM easier. If you implement against the byte code directly and the byte code changes then you have to make big changes in your compiler. By emitting Erlang terms you can make direct usage of the Erlang compiler (which should be more stable) in a later compilation stage. It's a similar situation for the LLVM and why they suggest you output LLVM IR and not bitcode.