SATB queue filtering involves two predicates: (1) does the object need to be marked, and (2) is the object marked.

Each time a queue is filtered, both predicates are applied to each entry. This means that if a post-filtered buffer doesn't exceed the completed buffer threshold, the remaining entries will be checked again by both predicates the next time the queue is filtered.

But the first predicate mostly doesn't change over time; an object is young, or it is old but allocated during this collection cycle, so can be filtered out. If it is an old object from before this collection cycle, then it will almost always remain so. The only exception is an old humongous object that gets eagerly reclaimed, rendering the queue entry "stale". That happens rarely, and will be caught later if not filtered out of the queue. So it might make sense to only apply the first predicate once per queue entry, and have the queue keep track of which entries have already been checked.