Robinhood's new L2 has one authorised transaction filterer, and its nonce is 6,092

The getter and the nonce answer different questions, and that gap is the whole finding. `getAllTransactionFilterers()` tells you one address is authorised to screen transactions on the chain ^[7]. The same address's transaction count tells you it has been used ^[10], and the destinations tell you what for: calls to the filter precompile at `0x...0074` with the selector for `addFilteredTransaction(bytes32)`, every one succeeding ^[11].

The mechanism is worth reading closely before pricing anything on this chain. ArbOS lets an authorised filterer register a transaction hash, after which the state transition function forcibly fails that transaction, including one force-included through L1 ^[9]. Force inclusion through the parent chain is the standard answer to a sequencer that will not take your transaction. Here it is not an answer, because the rejection happens a layer below the sequencer.

Timing narrows the read further. The first filter call landed 30 June 2026 at 14:53 UTC and the most recent the author saw was 9 August ^[12], which is 40 days for a counter sitting at 6,092: about 152 filtered transactions a day ^[18], consistent with the roughly 150 a day the author reports, alongside no published criteria, no volume disclosure and no appeals process he could find ^[13]. The first call also predates the 1 July public mainnet launch by a day ^[19], so the blocklist was warm before the chain was open.

The author is careful about what this does not show. The hashes reveal nothing about intent, and the reading establishes only that the capability is operational rather than dormant ^[14]. The filterer address carries no name or public tag, so attributing it to Robinhood is inference ^[15]. He also could not determine from the blocklist whether a contract deployment has ever been filtered ^[16], which is the question a deployer with revenue actually needs answered.

The method is the transferable part, and it is cheap. A "built on Arbitrum" badge is a marketing artefact; Nitro precompiles returning the single byte `0xfe` is a property the chain cannot fake without being one ^[3]. "Permissionless" is a word on a page; estimating gas for a contract creation from an address the chain has never seen either gets rejected by an allowlist or comes back with a price, and here it came back with a price ^[4]. A selector table is someone else's transcription; deriving both selectors from `keccak256(signature)` costs nothing and removes the transcription ^[8]. The author's summary of all three is to prefer evidence the subject cannot author ^[17].

Everything the documentation does claim held up: an Orbit L2 settling to Ethereum, ETH for gas at fractions of a gwei, permissionless deployment ^[1], chain ID 4663 ^[2], and the usual canonical infrastructure already deployed ^[5]. Governance is equally legible from a public read: one chain owner, an UpgradeExecutor proxy ^[6]. None of that is the risk. The risk is a screening path that no page describes, running at a rate nobody publishes, on a chain whose escape hatch does not cover it.