One honest miner preserves censorship resistance

Bitcoin's censorship resistance holds so long as a single economically rational miner with nonzero hash power will include a given transaction.

For (strongest version as argued): Exclusion requires unanimity among block producers; inclusion requires one defector, and fee revenue makes defection profitable by construction. As long as any miner anywhere will take the fee, a censored transaction confirms — later than usual, but inevitably. This is why relay filters and even majority miner-filtering cannot deliver true censorship: they only price-discriminate on latency.

Against (strongest version as argued): The one-honest-miner guarantee is weaker than it sounds. If the honest miner has x% of hash power, expected confirmation time is 10/x·100 minutes — with 1% honest hash power, roughly a week, which is functional censorship for time-sensitive protocols (Lightning penalty transactions, contract deadlines). Majority hash power can go further and orphan blocks containing the disfavored transaction, turning "one honest miner" into "one honest miner willing to lose money mining doomed blocks." And under a consensus-level restriction (as opposed to policy filtering), the honest miner's block is simply invalid — the guarantee evaporates entirely.

Nuance: The claim is a good approximation against uncoordinated, policy-level filtering and a poor one against majority-coordinated or consensus-level exclusion. Time-sensitivity of the use case determines whether delayed inclusion counts as resistance or as failure.

Common misstatements: "Censorship resistance means transactions can't be delayed" (delay is expected under partial censorship; permanence of exclusion is the bar). "One node is enough" (relay alone mines nothing; the guarantee needs hash power). "Miners must mine every valid transaction" (block construction is discretionary; that discretion is exactly the attack surface under debate).