Yes, it’s completely natural to apply some sort of future discounting, as something far away from “top blocks” are likely to:
- be trimmed
- be RBF’d
- timeout
so reflecting that uncertainty directly is something someone can explore further. I did some initial number analysis to see if it could help anti-pin(it can’t 
)
A minor note, but this also necessarily opens up to “mempoolfullrbf” style RBFs, good or bad.
One idea I’d had, which Suhas has essentially responded to above, is that we could:
- receive new chunk somehow that exceeds cluster limits
- allow temporary exceeding of limits
- re-linearize cluster with exceeded limits
- “chop off” the excess at the end
As Suhas notes, this can get pretty weird with pinning/anti-DoS. I don’t think it’s a dead-end, but more of a whiteboarding project at this point.