As another point, it seems like there are 2 distinct asks buried in this:
- stop looking for results when the pagination limit is hit, even while evaluating that all result nodes have every cascade field
- if the above isn’t easy then at least do not follow edges out of a level that would be cut by the pagination of that level (after it does get applied)
The latter alone would bring the hit UIDs of my example query to 100+6000+40=6140 instead of 100+300000+6000000=6300100.
But the first point would bring the total UIDs hit to 14, which is even better than 6140. 