A self-authored companion to Cadmus. Cadmus works by making the lens explicit — a declared codeword sitting at the top of the message. Most responses, most of the time, carry no codeword at all. That's not neutral. It's just unsigned.
Cadmus steers by declaration: critique + punch-up: [text] names, in plain sight, exactly which transformation is being asked for. That's the whole design — steering becomes visible and choosable instead of implicit.
But say nothing, prefix nothing, and a response still arrives — shaped by something. No codeword doesn't mean no lens. It means an unlabeled one: whatever a model's untriggered defaults happen to be, running exactly as hard as any declared codeword, just never named.
This page has been under a lens the entire time you've been reading it — a real one, doing real work, never declared to you as a codeword the way Cadmus declares its transforms. Toggle it off. Everything you're reading is still here; only the unnamed styling choices are gone.
currently: styled (the lens is on, undeclared, as it has been this whole time)
Cadmus's real contribution isn't adding a lens where there was none — there's always one. It's making the lens choosable instead of ambient, and named instead of assumed. The gap it closes is the gap between "shaped" and "shaped on purpose."