One of the things you can see at mempool.space view of future blocks and blocks in general is there seems to be a kinda "dynamic range" limit imposed somehow. Like, everything under a price is just no chance.
Something that would help stop this backlog of small transaction fees is if standard selection algorithms used a gaussian distribution instead of a band pass filter. The flow would slow but it wouldn't stop, I think this is a small thing to do that will cost virtually no difference but let the network progress more smoothly.
In this recent events I have seen trees of transactions get stopped up by one neglected fee rate being too low, and the sender abandoning it even after doing CFPP on it twice.
It briefly came within a day of the fee rate distribution today, but sadly it lingers in the distance, unable to complete until the rate drops under 20 long enough for it to get caught in a block.
I am a tail emission and very big denomination bits kinda guy also, but this case is a simple, cheap change that yields a better overall UX in congestion that doesn't shut the tx's that got dropped in by bad estimators at laggy rates. WDYT?