deposit(proof, extData, screening)for a TOKEN deposit;requestDeposit(asset, amount, screening)for an eTOKEN deposit. Verifying and finalizing that receipt need nothing more, because the receipt belongs to the depositor alone.
transact, the entry point for private transfers and withdrawals, refuses any positive extAmount with DepositRequiresScreening, so relayed jobs never carry a screening and the relayer never has to handle one. Screening sits at the entry to the anonymity set and nowhere else.
What a screening is
A screening is three fields — a randomnonce, an expiry expiresAt and a signature — and that signature is the pool’s screeningSigner’s EIP-712 signature of this message:
PrivacyPool, version 1, this chain id, the pool’s own address. Referral bonuses are signed in the same domain, under a different type.
Before the proof is checked and before anything is pulled, the pool recovers the signature with msg.sender as the depositor, and requires that
- the signer is the address the pool names as
screeningSignernow (otherwiseInvalidScreeningSignature, orScreeningSignerUnsetwhen the pool names nobody); - the screening has not expired, and does not expire more than
MAX_SCREENING_TTL— one hour — from now (ScreeningExpired); - its
noncehas not been spent yet (ScreeningUsed).
nonce (screeningUsed(nonce)) and logs DepositScreened(depositor, nonce).
So one screening is good once, from one wallet, into this pool, on this chain, under the signer the pool trusts. Nobody else can use yours, because the depositor is msg.sender rather than a field anyone chooses. It is deliberately not part of the zero-knowledge proof: swapping it for another screening of the same depositor changes nothing, and because a spent nonce is undone when the surrounding transaction reverts, a deposit that fails later leaves its screening unspent and reusable. Only a 65-byte signature in canonical (low-s) form counts.
The flow through the API
The browser never talks to the screening provider and never holds its credentials. It names its wallet to Fluton’s backend API:status: "screened" with the signed screening, status: "refused" with a reason code, or HTTP 503 when it could not get a clear answer at all. Anything short of a clear answer — a network error, a timeout, an HTTP error, a malformed body, an answer about another address or given under another configuration — becomes that 503, and nothing is signed. Screening never fails open.
A refusal says only that the address cannot deposit into the pool. What the screener found stays on the server, in the decision record the API keeps for every answer it gives.
The API does not ask CipherOwl for every request. It reuses CipherOwl’s latest answer about an address for up to a day by default, counted from the moment CipherOwl gave it, and concurrent requests about the same address share one call. A reused answer still buys a screening of its own, with a new nonce and the full validity. Reuse has a price: an address that CipherOwl flags after it was answered clear keeps getting screenings until that answer ages out. The API’s operator sets how long an answer is reused, and can cap the calls made in a day and in a month. Once a cap is spent, an address without a reusable answer gets the same 503 until earlier calls age out.
The operator can also answer for an address itself. An address on the operator’s denylist is always refused, and one on its allowlist always gets a screening. Neither is sent to CipherOwl. The allowlist is the operator’s own word that an address may deposit, so it carries the same trust as the signing key.
The SDK asks first, before rebuilding the tree, proving anything or approving any token, so a refused wallet learns it at once and spends nothing on gas or proving time. It then checks who signed the screening against the pool’s current screeningSigner, so a screening service whose key the pool does not trust is caught there too, rather than by a reverted deposit. A screening is valid for 30 minutes by default. Just before sending, the SDK fetches a new one if less than a minute of validity is left by the chain’s own clock, and once more if the pool calls the screening stale during simulation. Stale means expired, or spent in the meantime. A refusal is a different thing: it is shown as screening’s answer rather than as a failure, and it is final for that wallet.
A refusal is about the depositing address, not about the amount or the token. The only way past it
is a wallet that screening accepts.
Trust and failure modes
The pool takes the screening signer’s verdict as it is, so screening is as strong as two keys: the signer’s, and the owner’s, which names it withsetScreeningSigner. The signer’s key never sends a transaction and cannot move funds, but whoever holds it can let any address deposit. The owner cannot switch screening off: the signer may never be the zero address. Naming a new signer refuses, at once, every screening the old one signed. A pool upgraded from code that predates screening names nobody and refuses every deposit until its owner names a signer.
If the API or CipherOwl is unreachable, deposits stop and nothing else does. Transfers, withdrawals, verifyDeposit and finalizeDeposit never need a screening, so money already inside the pool keeps moving. And because screening applies at entry only, a depositor flagged after depositing is already in the anonymity set.
What the screener learns is what a deposit makes public anyway: the depositor’s address, the pool and the chain. Each screening a deposit spends is logged on-chain as DepositScreened, naming the depositor and the nonce, and the API records every decision (screened, refused or unavailable), together with the screening it signed and whether it asked CipherOwl, reused an earlier answer or took the answer from one of its lists, in its own database for the audit trail. What stays public puts that next to everything else a deposit reveals.
Screening is the first thing the deposit flow does, so it shows up in every deposit path: see TOKEN and eTOKEN for the two-step eTOKEN route it guards and Contracts for the exact signatures and errors.