Card billing hides authorization, tokenisation, retries and cancellation rules behind a payment provider. Stablecoin products should not reproduce that opacity. A wallet is not a card on file, an allowance is not unlimited consent, and one signature should not silently become permanent authority to move funds. A sound subscription product makes consent a first-class object: who may charge, which asset, on what network, up to what amount, at what cadence, until when, and how the customer revokes it.
Four decisions before code
- Separate the commercial plan, the payment mandate and each executed transfer.
- Default to bounded amount, cadence and duration rather than unlimited approval.
- Revocation and reduction are primary product paths, not support-only links.
- Support smart accounts and push renewals instead of assuming every wallet can sign a permit.
1. Use three objects instead of a stored payment method
The plan describes the commercial promise: product, price, billing period, trial, benefits and change rules. The mandate records payer consent: wallet, network, token contract, authorised recipient, cap per cycle, aggregate cap, start date and expiry. A charge is one execution attempt. The three must be linked, but they should never be collapsed into one mutable record.
Separation lets a merchant propose a new price without rewriting old consent, pause a failed execution without cancelling the plan, and close a mandate without deleting service history. Each object needs its own identifier, version and state. Support can then distinguish a commercial dispute from an expired authorization, insufficient balance or a transfer that has not reached operational finality.
- Plan — what the merchant offers and at which price.
- Mandate — what the customer allows and within which boundaries.
- Charge — one billing attempt with a specific time and amount.
- Entitlement — the product access granted after a payment decision.
2. No single execution model works for every wallet
A product can use customer-initiated transfers each cycle; allowance plus transferFrom; ERC‑2612 permit to set an allowance through an EIP‑712 signature; ERC‑3009 transfer authorization; or a smart-account module that performs scheduled transfers within policy. Each model makes a different trade-off among convenience, compatibility and standing authority.
ERC‑2612 is not a subscription standard. It adds a signed way to set an ERC‑20 allowance with a nonce and deadline, but the resulting authority remains an allowance. ERC‑3009 authorizes transfers rather than defining a recurring commercial agreement. Draft proposals such as ERC‑6932 for subscription tokens and ERC‑8255 for expiring approvals are useful design evidence, but draft status is not an interoperability guarantee. ZeroFee therefore needs a capability registry and fallback paths rather than one universal assumption.
3. Consent must be legible, bounded and provable
The approval screen should state the merchant, plan, asset and token contract, network, maximum charge, cadence, aggregate cap, expiration, price-change policy and revocation path. EIP‑712 typed data can make a signature more intelligible than opaque hex, but EIP‑712 does not provide replay protection by itself. The system needs domain separation, a nonce, a deadline and mandate state enforced by the server or contract.
A material change requires new consent. A price increase, token replacement, network migration or spender change is not a silent settings update. The product may offer a migration and explain it, but it must not expand existing authority in the name of convenience. Every charge should preserve the terms version and mandate version that justified it.
4. Bound the damage before optimising friction
Unlimited approval reduces repeat signatures but creates durable authority that may outlive the user's memory of the service or survive a spender upgrade. A subscription should prefer a per-cycle cap, aggregate cap and lifetime. If a plan costs 20 USDC per month, permanent unlimited authority has no automatic justification. A small disclosed buffer can cover tax, usage or an agreed change without making the mandate open-ended.
Smart accounts can use allowance or standing-order modules with destination allowlists, token limits and periodic resets. Safe documentation notes that modules can enable recurring transactions and spending caps, while warning that a malicious module can take over the account. Treat a module as production infrastructure: audit it, restrict it, test emergency stop and removal, and roll it out with limited value.
5. A scheduler is not business truth
A cron job can open a charge at the expected time, but it cannot prove that execution is permitted. Before each attempt, verify that the plan is active, the mandate is current, the amount is within its cap, asset and network match, the cycle has not already been paid, and the idempotency key is unique. Record proposed, authorized, submitted, observed, settled and failed transitions as an append-only state machine.
Block time is not a commercial calendar. Monthly may mean the tenth day of each month, not every 30 days. Time zones, month-end and daylight-saving transitions need explicit rules. Grace periods are also product decisions: whether access remains active during retry, when the customer is notified and how many attempts are allowed before pause. Customers should know that policy in advance rather than discover it through a surprise charge.
6. A failed charge is a conversation, not a retry loop
A charge can fail because of balance, allowance, an expired mandate, reused nonce, gas, RPC, network availability or a different token contract. Each cause demands a different response. Repeating transferFrom after consent was revoked is not recovery; asking for a new signature when only the observer is delayed is unnecessary friction.
Dunning should distinguish a temporary technical failure, required customer action and deliberate cancellation. The message explains what is missing and what happens to access without claiming that a wallet was simply declined. Retries need a ceiling and backoff, and every attempt remains attached to one billing cycle to prevent double collection. A late settlement should restore an entitlement exactly once.
7. Cancellation must close authority as well as service
Cancel subscription should stop future attempts, close the internal mandate and provide a route to revoke an onchain allowance or module where one exists. A merchant often cannot revoke on the customer's behalf. In that case, the product should show the action link, spender address, network and verified status. The interface must distinguish stopping service from removing contract authority.
Entitlement remains a separate object. A customer may cancel yet retain access until period end; a failed charge may be inside a grace window; or a late transfer may arrive after access closed. A shared state machine for subscription and payment prevents a chain observer from opening access without understanding the commercial agreement.
8. Ledgers, reconciliation and product metrics complete the system
Every charge links to an invoice or billing period, mandate, transaction and double-entry postings. The chain proves movement; the ledger explains revenue, tax, platform fees and refund liabilities. Reconciliation looks for duplicate collection, missing cycles, wrong assets, late payments and entitlements that disagree with settlement. Exceptions receive owners and service levels rather than disappearing into manual edits.
The useful metrics extend beyond MRR. Track completed-consent rate, the share of time-bounded mandates, revocation time, failure rate by cause, recovery without double charging, charges blocked by caps and customer comprehension of the next amount. Dor Arad's product principle is that good automation does not move money quietly. It executes exactly what was agreed, explains every charge and stops when the boundary ends.
Sources and further reading
Official sources were reviewed on November 10, 2026. ERC‑6932 and ERC‑8255 are explicitly identified as drafts. This is product and engineering analysis, not legal or financial advice.
- ERC‑20: Token Standardcreated November 19, 2015
- ERC‑2612: Permit Extension for EIP‑20 Signed Approvalscreated April 13, 2020
- EIP‑712: Typed Structured Data Hashing and Signingcreated September 12, 2017
- ERC‑3009: Transfer With Authorizationcreated September 28, 2020
- Safe Docs — Smart Account Modulesreviewed November 10, 2026
- ERC‑8255: Expiring Token Approvals [DRAFT]draft, May 2026
- ERC‑6932: Subscription-Based Token [DRAFT]draft, April 25, 2023
ZeroFee Pay is an independent Dor Arad project in development and sandbox mode. Do not move real assets before the relevant legal structure, licensed providers, security review and operational controls are complete.
