> “if X percentage of filtering nodes are unable to 100% reliably stop spam, then they don’t work” is just a fallacious claim. The obvious contradiction is that if 99.9% nodes aggressively filter spam, and filtered txs still end up in blocks (they can), then filters also can’t be used for censorship (the whole argument falls apart)

So we agree: censoring doesn't work if it's not at the level of consensus: if it did, bitcoin would fail. but further, what kind of restrictions can and should exist at the consensus level is very controversial. even limiting the *rate* of transactions per block was controversial enough to create a full blown war in Bitcoin, for years. The only argument that falls apart is the one saying there's any point in doing mempool filtering; while there are very clear and concrete negative outcomes (though I wouldn't overplay them) from filtering, too.

> And no, filters work not because they reliably prevent spam from getting into blocks, but because they force behaviour that makes spam economically and rationally self-defeating (to a large extent)

I don't agree. Your argument seems to be that they make spam much more expensive, but I see absolutely no reason to think that, specifically because, as we already agreed, they don't work - the same transactions you are filtering end up in blocks, relayed directly to miners, often.