I second this request. In particular, please clarify whether "preference" and "logical correctness" are presented here as examples of "explicit considerations". And whether whole agent should be parsed as including all the sub-agents? Or perhaps as extrapolated agent?