The sourcing autopilot is the fast, cheap part of the vault. Twice a day it goes looking for roles, checks each one is still live at its source, and grades it against fixed criteria before dropping the shortlist into the vault. Because it runs against the real postings rather than a stale cache, a role that has closed is caught by the check rather than drafted against in vain.
Left to itself a loop like this optimises the number it can measure, roles found, and that number is not the one that matters. So the autopilot is deliberately capped. When more than a handful of packs are already waiting on a human decision, drafting freezes by contract. The reasoning is a bottleneck argument: adding more graded roles behind a full review queue does not speed anything up, it just grows a backlog and hides the real constraint, which is a person's attention. The fast loop is not allowed to run ahead of the slow one that owns its purpose.
That cap is the whole discipline in one rule. A search that cannot say whether a role is worth applying to must not be the thing that decides how much work exists.