The useful public distinction between card kinds is whether the originating work thread must wait. A blocking card parks that thread until an answer returns. A non-blocking card can be read without making that thread wait.
This is a lifecycle effect, not a measure of how important the card feels. ROOK does not ask its mechanical driver to infer urgency or reorder work from the wording.
[Visual 2.2 — A two-column matrix labelled "Blocking" and "Non-blocking". The rows are "originating thread", "other runnable work" and "operator attention". Blocking shows "parks", "can continue" and "answer needed to proceed". Non-blocking shows "does not park", "can continue" and "visible without halting the thread". A note under both columns reads "the coding CLI authors the card; the driver applies only the recorded effect".]
Blocking cards
A blocking card means the coding CLI cannot responsibly continue that one thread without the operator's intent. ROOK records the card and parks the thread. The driver can still start another session when the coding CLI identifies different work as runnable.
When the answer arrives, it is recorded against the card and a fresh coding session can return to the parked thread. If no other task is runnable while the card is open, the work loop waits.
Non-blocking cards
A non-blocking card leaves its originating thread free to continue. It can carry information the operator should see without pretending that every update is a decision boundary. The card still belongs in the record and on the surfaces, but it does not create a parked thread merely by being visible.
The two cases can therefore share one card shape while having different effects on work. That effect is recorded explicitly; it is not guessed from tone, length or a headline.
Limits
This page describes the stable blocking distinction, not ROOK's internal catalogue of attention labels. Blocking does not mean globally urgent, and non-blocking does not mean unimportant. Nor does a blocking card stop every project or device: it parks its own thread, while continuation depends on whether the coding CLI can identify other runnable work.
Next
Follow the operator's choice back to its source in Answering.