The holder count and the execution-time recheck both read the directory export, so both inherit how that export treats a disabled account. If a leaver's account is disabled but keeps its group memberships in the export, it still counts as a holder, and an approval that person signed shortly before leaving still passes the second check when the destroy runs.
I'd put account state in the data the policy reads: either the export drops disabled users, or User carries an active attribute that destructive-needs-approval requires and the holder count filters on. Then the policy, the recheck and the count agree on who can approve, and reactivating a dormant account shows up in the diff as a move from one holder to two instead of passing silently.
dvaddi