Bitcoin Pool Censorship Scorecard: Who Filters Transactions — With Evidence

“Which pools censor transactions?” is usually answered with vibes — jurisdiction guesses, marketing copy, and a decade of rumor. This scorecard answers it with evidence: a pool only gets an observed filtering verdict when there is a dated public incident that survived false-positive screening, and everything weaker is labeled inferred or unknown, never dressed up as fact. It covers sanctioned-address filtering, OP_RETURN and inscription data policy, KYC, who controls the block template — and the escape hatch a miner can actually use today.

How filtering is actually detected

The instrument of record is miningpool.observer (open source, by Bitcoin developer 0xB10C): it continuously builds its own block template with a well-connected Bitcoin Core node and diffs it against every mined block, flagging transactions a profit-maximizing pool should have included but didn’t. A flag is not a conviction — young transactions, fee displacement and block-weight reservation produce false positives, which are screened by hand before anything is published. That screening is why Foundry, ViaBTC and SBI Crypto sit at none-observed on this scorecard despite each having had a flagged block, while F2Pool’s repeated misses — acknowledged by its own co-founder in November 2023 and re-flagged for December 2024 blocks — are the real thing.

The two kinds of “filtering” are not the same thing

Sanctioned-address filtering (F2Pool’s patch, MARA’s 2021 “clean blocks” era) drops transactions because of who is transacting — that is censorship in the political sense, and the record shows it fails: every filtered transaction confirmed anyway in another pool’s block, and MARA reversed course within a month under community pressure. Data-policy filtering (OCEAN’s Knots-based default template) drops transactions because of what they carry — inscription data, oversized OP_RETURN payloads — and OCEAN is open about it. The 2025 Bitcoin Core v30 datacarrier change turned that stance into the network’s loudest policy fight; our Core vs Knots relay-policy guide covers how the same split plays out at the node level. The scorecard records both kinds, labeled for what they are.

The structural fix: build the template yourself

Any pool-built template — filtered or not — is a policy you inherit. The durable answer is miner-controlled templates: OCEAN’s DATUM protocol has been live since September 2024 and lets each miner build blocks from their own node while keeping pooled payouts; Stratum V2’s job-declaration role promises the same but is not a generally available customer option on major pools yet (our Stratum V1 vs V2 matrix tracks honest deployment status); and solo stacks like a self-hosted Public Pool or ckpool instance give a Bitaxe or full-size fleet complete template sovereignty. For fees, payout schemes and decentralization scores across all these pools, see the mining pool database and the live pool centralization tracker; for the bigger picture, start at the mining pools hub.