All three are the shipped code — card, tokens, wordmark and rows read out of the built widget, so what you tap here is what the connector draws. Try the rubric: move something across the line and watch what changes.
Why a line and not just an order. A rubric arrives with six to eight things, and an interview can only really probe three or four. An ordered list says which is first; it does not say which ones get asked about at all. The line does, and moving a row across it is the only place anything changes — so the decision and the feedback are the same gesture.
What a config picks, and what it does not. Claude sends the criteria, their sub-points, and whether the line sits after three or four. It cannot reword the line: two rubrics have to mean the same thing to be worth comparing.
Still to do before merge. The rubric widget in the connector still renders the old importance sliders — this control is built and not yet wired in. Nothing here is merged.