Moving a token from one address to another is a technical operation. A payout is a business promise. A marketplace must know who earned value, which transaction created the entitlement, when the balance becomes available, who may change the destination, which network is supported, what happens when execution fails and how the obligation is proven closed. Stablecoins can shorten the path and expand global reach when those questions form one coherent product. Starting with transfer merely postpones the discovery that a blockchain does not manage customers, liabilities, approvals or books.
The thesis in brief
- Payout entitlement is created in the business ledger; an on-chain transaction settles it but does not define it.
- A payee is an identity, verified destination, terms and policy—not just an address.
- Available balance, payout request and transfer attempt are separate objects with different lifecycles.
- Batching, sponsorship and scheduling are product and risk decisions, not only gas optimizations.
- Travel Rule and jurisdiction differences require a policy layer before rail selection.
1. Start with the liability, not the wallet
In a marketplace, customers pay, sellers or creators earn, the platform charges a fee and sometimes holds a reserve. None of those relationships exists in ERC-20. The token standard can move value and emit a Transfer event, but it does not know whether an order was fulfilled, a refund window closed or a seller may withdraw. The first source of truth is therefore a ledger obligation: who owes whom, in which accounting currency, because of which event and under what conditions.
The event that creates entitlement must be explicit. Confirmed delivery, a completed milestone, a closed subscription period or a released hold produces balanced postings. Funds may sit in one custody address while each recipient’s balance remains a ledger result, not a balanceOf query. One obligation can then be frozen, adjusted or routed to another network without mixing the rights of other participants.
From a product perspective, payout begins when value moves from pending to available. The transition records a policy version, timestamp and reason. The interface should show what can be withdrawn now, what is expected to become available and what is held for review. That clarity matters more than a fast Withdraw button because it prevents users from seeing a balance the platform cannot actually release.
2. A payee is a managed entity
A cryptographically valid address is not necessarily the correct destination. A marketplace needs a payout profile joining the business entity, account owner, KYC or KYB status where applicable, operating region, preferred asset and network, verified address and change policy. The profile is versioned. A payout created today must remember which destination and status were active when approval occurred.
Adding a destination is a sensitive operation. The product can request a wallet signature, issue a one-time challenge and alert the user through a separate channel. A destination change after suspicious activity may enter a cooling period or require additional approval. Login alone is insufficient because a stolen session should not immediately become a stolen balance. Separating account access from destination management limits the failure surface.
The same user may choose USDC on Base for routine payouts and USDC on Ethereum for a large amount. These are distinct rails with different contracts, gas, finality and operational risks. Circle’s documentation explicitly models support as a chain-and-currency pair and warns against unsupported assets or routes. The product catalog should lock the approved pair rather than trust a symbol supplied by the client.
3. Balance, request and execution attempt are separate objects
Available balance describes an economic right. A payout request describes an instruction to use part of that right. A transfer attempt describes a technical effort to settle the request. Combining them creates serious faults: an RPC retry can become a duplicate payment, a failed transaction can erase entitlement, and a destination edit can retroactively change an approved request.
A payout request should contain an idempotency identifier, payee, amount, accounting currency, asset, network, destination snapshot, fee policy, source of funds and acceptance of terms. Creation moves value from available to payout pending in the ledger. Cancellation before execution posts a reversal back to available. A mutable balance field should never be the only record of that movement.
Every attempt records a nonce or transaction hash, status, provider, time, gas and classified error. A replacement transaction is not a new payout; it is another attempt for the same obligation. A new request with another idempotency key must pass the balance check again. This structure supports safe retry and gives customer support a precise answer about what happened and what is still owed.
4. A state machine prevents duplicate payment
A practical path can include requested, eligibility_review, approved, scheduled, signing, broadcast, confirming, completed, failed, cancelled and exception. Every transition has an authorized actor and a condition. Support may open a request but not approve a high amount. A policy engine may approve a small payout automatically. Treasury broadcasts only approved work, and the observer marks completed only after a valid receipt and the required finality.
The execution worker needs a logical lock on the request, not only a process lock. If two workers receive the same message, both must reach one outcome without broadcasting twice. An outbox pattern connects the ledger commit to creation of the execution task. Idempotency survives both an external provider and the internal event bus. A redelivered webhook is not permission to create another posting.
Failure is not one terminal state. Insufficient gas, RPC timeout, blocked destination, changed policy and contract revert require different responses. Only a classified final failure returns funds to available. A timeout after broadcast requires a transaction lookup before any retry. The rule is simple and critical: technical uncertainty is not permission to pay again.
5. Authorization should surround actions, not a generic role
OpenZeppelin notes that production systems tend to outgrow one owner and need roles, targets and functions. A marketplace should separate proposer, approver, executor, destination manager and reconciler. One person or service should not be able to replace a destination, approve its payout and sign the transaction alone.
Authority should be narrow: who may approve how much, for which tenant, asset, network and time window. An unusual operation receives a delay or a guardian able to cancel it. A smart account can enforce a threshold and allowlist on-chain, while business policy also remains off-chain because it depends on KYC, reserve, disputes and data that should not be public.
API access is not economic authority. A merchant key may create a payout request, but only an internal policy decision turns it into approved. Signing occurs in an isolated execution environment, with secrets unavailable to the application or an agent prompt. Every authorization decision stores its policy version so an investigation can reconstruct why the action was allowed.
6. Batching reduces cost and concentrates risk
When hundreds of recipients are waiting, batching can reduce gas and operational cost. A large batch also increases blast radius: a wrong destination, decimals error or flawed contract can affect many obligations together. Before implementation, decide whether the batch is atomic, whether one failing item reverts all items and how every transfer maps back to one payout request.
A good scheduler groups by network, token, time window and risk class, not only by clock. Urgent or high-value payouts may use a separate rail. Caps on total value and item count bound loss. A deterministic preview calculates recipients, amounts, fees and required balance. A hash of that manifest is stored before approval so execution cannot change the batch after reviewers consent.
Gas sponsorship is also a product decision. If the platform pays, it needs a budget, abuse controls and cost attribution. If the recipient pays, the payout is not truly frictionless. The product may absorb cost below a threshold, deduct a transparent fee or offer a lower-cost cadence. Economics should not disappear inside an unexplained spread.
7. Availability and finality are two different decisions
A seller’s entitlement may become available before the original incoming payment reaches full finality, but the platform is then extending credit. ZeroFee Pay’s article on finality distinguishes observed, accepted and settled. The same distinction must govern payout. Value should not leave the platform based on a status designed only to improve checkout experience.
Before payout, verify that the funding sources are settled under policy, reserve remains intact and no dispute is open. If early payout is a feature, define a separate exposure, seller limit, credit price and legally workable clawback path. A clear interface label prevents credit from being presented as money already settled.
After broadcast, completed is not determined by a green wallet check. It depends on the receipt, token event, confirmations and network policy. Different networks require different adapters. A recipient webhook comes from the state machine, while the API remains the source of truth when delivery fails.
8. Travel Rule and compliance come before rail selection
FATF’s July 16, 2026 report found that 83% of survey respondents had passed legislation implementing the Travel Rule, up from 73% in 2025. The rule requires relevant VASPs and financial institutions to obtain, hold and transmit specified originator and beneficiary information for covered transfers. This does not mean every payout is identical or every marketplace is a VASP. It means legal classification and jurisdictions determine which data and providers are required.
A policy engine should run before transaction creation. It evaluates jurisdiction, recipient type, amount, asset, hosted or unhosted wallet, screening, information requirements and provider limits. The allow, review or deny decision is stored with an explainable reason code. Personal information should not be placed on-chain; the communication and evidence layer remains off-chain and appropriately protected.
Obligations require legal advice suited to the product structure. Sound engineering does not assume one legal conclusion for every market. It supports configuration by country, minimizes stored information and separates compliance evidence from general product data. A regulatory change can then update policy without rewriting the payment rail.
9. Reconciliation closes the obligation
A blockchain proves that a transfer executed; the ledger explains why. Reconciliation compares the approved payout request, attempts, chain events, postings and treasury balances. Completed requires the asset, network, contract, destination and amount to agree. A transaction may succeed technically while sending the wrong token. In that case, the business liability is not closed.
A daily process surfaces approved-but-unbroadcast work, stuck transactions, completed attempts without postings, postings without evidence, fee variance and custody balances that do not reconcile. Every exception has an owner and SLA. Rebuilding the indexer from the same block should produce the same result, and decoder versions should be retained so a code change cannot silently rewrite history.
Recipient statements use the same truth. A statement shows opening balance, earnings, holds, fees, payouts and closing balance, with a transaction link where available. Support, finance and the user then discuss the same identifiers. Transparency is not only user experience; it reduces disputes and manual operations.
10. A build order that produces a reliable product
Start with the entitlement and balance ledger, then create the payout profile and verified destination, followed by the idempotent request and state machine. Add policy and authorization, an executor for one network and asset, an observer, reconciliation and a recipient statement. Only after every exception can be explained should the product add batching, another network or early payout.
At every stage, test failure: double click, provider timeout after broadcast, lookalike token, changed destination, insufficient gas, duplicate worker, stricter policy mid-run, reorganization and a batch with one failed item. Success rate is not enough. Measure available-to-completed time, review rate, payout cost, prevented duplication, open exceptions and reconciliation time.
Dor Arad’s principle for ZeroFee Pay is that reducing friction does not require hiding complexity. The product should organize it: show recipients what they earned and when, give platforms bounded control, and give finance balanced evidence. A payout then becomes more than a transfer command. It becomes a distribution rail the business can measure, explain and expand.
Sources and further reading
Primary sources were checked on November 17, 2026. Regulatory scope depends on the product and jurisdiction and requires legal verification.
- EIP-20 — ERC-20 Token StandardNovember 19, 2015
- Circle — Supported Chains and CurrenciesChecked November 17, 2026
- OpenZeppelin — Access ControlChecked November 17, 2026
- FATF — Seventh Targeted Update on Virtual Assets and VASPsJuly 16, 2026
- ZeroFee Pay — Stablecoin Treasury ArchitectureRelated reading
- Dor Arad — ZeroFee Pay projectProject context
ZeroFee Pay is an independent project by Dor Arad. This article explains product and engineering choices; implementation requires security review, accounting controls and legal advice appropriate to the product and its jurisdictions.
