So actually, I just reworded the article to avoid saying "the mempool" 😂
Comment to nostr:npub1ej493cmun8y9h3082spg5uvt63jgtewneve526g7e2urca2afrxqm3ndrm on his article about one-shot RBF:
Section 2, rule 2 and 3: while it is entirely logical that part of the ruleset is some measurement of expectation of confirmation of the replacement, I'm wondering, is there a fly in the ointment as mentioned several other places in the article: you are here referring to *the* mempool, which can only be the local node's mempool. Since almost none of those mempools are the mempool of the node that mines the next block, could that cause a problem? (Ironically, censorship such as seen in Knots/Ocean's policy (i.e. actually meaningful censorship of economically significant transactions) might mean that a node sees N=1 when in "real life" it's more like N=20 depth for "the" uncensored mempool).
I'm guessing the answer is "no, not a problem" because from a local perspective, one-shottedness is achieved, and it's only locally that you need/want that?
Hmm, anyway, I actually do kind of like the one-shot idea; as you pontificate near the end, it's debatable whether the multi-shot behaviour of the simplest possible rule (x% higher fee rate) is really a problem or not, but it's nice to remove it.
https://petertodd.org/2024/one-shot-replace-by-fee-rate#fnref:mempool-consensus
Discussion
As for measurement of % chance of confirmation, I think that should track very well with feerate compared to the rest of the mempool. This stuff doesn't have to be perfect. Just good enough.
Another way to do that, with a potentially simpler implementation, would be to allow replace-by-fee-rate after some timeout. Eg after # blocks have been mined with the tx remaining in your mempool.