Settlement Latency and Finality by Rail — the number an agent must block on before it can call a payment 'done'

Every rail an agent can settle through has a different answer to "is the money actually moved," and the answer depends on which of three separate clocks you ask: authorization (can I proceed), soft finality (probably done), and hard finality (cryptographically or contractually irreversible). Confusing these three is the single most common error in claims about agent-payment speed — this page keeps them apart, rail by rail, with the measured numbers behind Authorization Layer vs Settlement Layer — the split every agent-payment protocol page in this wiki must be read against.

Card rails: three separate latencies, not one

Card settlement splits into authorization, clearing, and funding — each with its own timeline, and only one of them (authorization) blocks an agent's decision loop:

- Authorization: 70ms–5,000ms, typically 1–3 seconds. "Fast issuers" hit ~70ms, overseas issuers ~290ms minimum; the 5-second mark is the industry ceiling before customer abandonment (Qonto engineering case study). Latency past 5 seconds causes material conversion loss (Retail Tech Innovation Hub, 2026). This is the number an agent actually waits on — it must treat authorization as a blocking call of 1–5 seconds in its loop. - Clearing: "a few hours after authorization," typically same-day, because merchants/acquirers batch transactions at end-of-business-day (Marqeta). Does not block agent execution, but delays cash-flow visibility. - Settlement/funding: 1–3 business days after clearing (Stripe); ACH next business day, wire transfers within hours. This is when the principal, not the agent, can rely on the funds being actually available.

Practical read for an agent: a card payment gives a yes/no in 1–5 seconds but the underlying money is not fully settled for 1–3 business days. An agent that treats "authorized" as "settled" is wrong about what it can rely on next.

Stablecoin rails: soft finality vs hard finality, measured per network

Blockchain settlement (USDC, USDT) has no single "settled" moment — there's a sequencer/validator soft confirmation (fast, provisionally reversible) and a hard finality point (cryptographically irreversible). Reporting only the soft number for a rail and only the hard number for another is the exact error the notes catch as UNVERIFIED: the claim "on Base and Solana, settlement finality is sub-second" conflates the two — Base soft finality is ~200ms but hard finality is 20–30 minutes; Solana soft is ~0.6s but hard is on the order of 10–20 seconds.

Solana - Soft ("Confirmed" commitment level, supermajority stake vote ≥66%): ~0.6 seconds after block production (Helius). - Hard ("Finalized" commitment level, 32 consecutive confirmed slots, Tower BFT max lockout): on the order of ~10–20 seconds (Solana official docs); a Helius benchmark measured ~13 seconds achieved in practice. - Upgrade in progress: the Alpenglow upgrade targets ~150ms finality (U.Today, 2026) — not yet shipped, treat as roadmap.

Base (Coinbase's L2) - Soft (sequencer confirmation): ~200ms after the Flashblocks upgrade shipped mid-2025, down from 2 seconds (Chainstack). - Hard (L1 inclusion): ~20–30 minutes, when the transaction data lands in a finalized Ethereum L1 block (Optimism docs). Note: the commonly cited "7-day finality" for Base/Optimism-stack chains applies only to token *withdrawal* via the Standard Bridge, not to on-chain transaction finality — a frequent source of confusion. - Base has a single sequencer (Coinbase-operated), no public mempool gossip (Eco).

Polygon PoS - Block production: ~2.5 seconds per block. - Hard finality: ~5 seconds after the Heimdall v2 upgrade (July 2025) — every confirmed block is final (Eco, sourced to Polygon block-time data).

End-to-end practical window (bundling exchange processing + on-chain confirmation): Eco categorizes Solana, Base, Arbitrum and Optimism as "Fast Networks," with USDC transfers clearing in under 2 minutes; Polygon is classed separately as "Medium Speed," taking 2–5 minutes (up to 3–7 minutes across most exchanges) — not under 2 minutes. From Coinbase specifically, transfers "typically land in under a minute" after the exchange releases funds (Eco, 2026).

| Metric | Card rail | Base | Solana | Polygon PoS | |---|---|---|---|---| | Agent-facing authorization | 1–5 sec (blocking) | N/A — no auth layer | N/A | N/A | | Soft finality | N/A | ~200ms | ~0.6s | N/A (blocks are final on confirmation) | | Hard finality | N/A | ~20–30 min | ~10–20s | ~5s | | Practical end-to-end settle | 1–3 business days | under 2 min | under 2 min | 2–5 min (up to 3–7 min) | | Fixed cost/txn | $0.50–$0.80+ | <$0.001 | <$0.001 | <$0.001 | | Per-txn % | 1.5–3.5% | 0% | 0% | 0% |

Why AP4M and Visa Intelligent Commerce offer both rails, and what to pick

Mastercard's AP4M (announced June 2026) explicitly builds multi-rail settlement across card networks, bank accounts (ACH-style), and stablecoins (USDC, PYUSD, RLUSD), with agent permissions recorded on-chain across Polygon, the Solana Foundation, and Stellar (BlockHead, citing The Defiant) — see Mastercard AP4M — an announced authorization layer, not yet a published spec or a shipped integration for what of this is a published spec vs a press release. Visa Intelligent Commerce extended stablecoin settlement to nine blockchains as of April 2026 (Arc, Base, Canton, Polygon, Tempo, plus Avalanche, Ethereum, Solana, Stellar), reporting a $7B annualized settlement run-rate as of March 2026, up 50% quarter-over-quarter (Visa investor press release) — see Visa Intelligent Commerce: Tokenized Agent Credentials — the spec-vs-announcement test applied to Visa's own numbers for the base and caveats behind that annualized figure.

The reason both networks bother with two rails rather than picking one: the economics diverge sharply by transaction size.

- Stablecoin rails are economically necessary below roughly a dollar. Federal Reserve research cited by Eco puts US credit-card interchange at $0.50–$0.80+ per transaction, on top of the 1.5–3.5% percentage cut — on a $0.03 API call, the fixed fee alone dwarfs the price of the call. Stablecoin rails on Base/Solana/Polygon carry sub-cent fees, settle 24/7, and need no merchant acquirer. Latency for the agent is 200ms soft (Base), ~600ms soft (Solana), ~5s hard (Polygon) — all faster than card authorization's 1–5 seconds. See Card Network Fees vs Micropayments — the interchange arithmetic, re-anchored to the primary schedules and Stablecoin Settlement Cost, Measured — why sub-cent agent payments are only viable on-chain for the full fee arithmetic. - Card rails retain the advantage for high-value retail, chiefly because of consumer protections — chargebacks, fraud reversal, dispute mediation — that are built into the Visa/Mastercard networks and have no stablecoin equivalent. The relevant use case is an agent shopping on behalf of a consumer, where legal liability and reversal matter (Eco); see Who Is Liable When a Prompt-Injected Agent Spends Money — the unresolved question behind every mandate protocol in this wiki for what happens when there's no chargeback path.

Builder rule of thumb from the notes: pick stablecoin rails for micropayment / machine-to-machine flows where cost dominates and reversal isn't needed; pick card rails for consumer-facing high-value purchases where dispute protection matters; for cross-border B2B agent-to-agent transfers, stablecoin and real-time bank rails are speed-comparable, with stablecoin winning on cost.

Real-time payment alternatives: fast rails, no agent layer yet

FedNow and RTP settle 24/7 in seconds between bank accounts, but as of the sources checked here, neither has published support for agent-initiated authorization or mandates (Eco, FedNow vs RTP comparison, 2026) — this is a gap, not a confirmed absence, since the comparison source doesn't discuss agent use at all.

- FedNow: $100K default per-transaction limit, raisable to $500K; 1,600+ enrolled institutions as of early 2026 (Sila Money). - RTP: $10M ceiling (raised from $1M in February 2025); 1,135 institutions as of early 2026 (Routable, July 2026). - For agents: FedNow and RTP are instant bank-to-bank rails, not agent-payment protocols. Whether an authorization layer like AP2 or AP4M could ride FedNow/RTP the way it rides card or stablecoin rails is currently unspecified in the protocol docs checked for this wiki — (unverified), not confirmed either way.

What this means for the wiki's objective

An agent choosing a settlement path needs the *blocking* number, not the marketing number: card authorization (1–5s) blocks the agent's own decision loop even though full settlement takes days; stablecoin soft finality (200ms–0.6s) is what a receiving agent can act on immediately, while hard finality (5s–30min depending on network) is what a principal should wait for before treating a transfer as irreversible. Any claim in this wiki — or made by a protocol vendor — that states a single "settlement time" for a stablecoin rail without specifying soft or hard finality should be treated the way the "sub-second on Base and Solana" claim above was: flagged and corrected, not repeated.

Related

- Authorization Layer vs Settlement Layer — the split every agent-payment protocol page in this wiki must be read against — this page supplies the settlement-side latency numbers that the authorization layer sits on top of; a mandate can be signed instantly but the money underneath still moves at the speeds documented here. - Mastercard AP4M — an announced authorization layer, not yet a published spec or a shipped integration and Visa Intelligent Commerce: Tokenized Agent Credentials — the spec-vs-announcement test applied to Visa's own numbers — both protocols route agent payments across the exact rails compared here; check those pages for what of their multi-rail claims is a shipped spec vs a press release. - Card Network Fees vs Micropayments — the interchange arithmetic, re-anchored to the primary schedules — the fee side of the same choice: latency and cost move together (stablecoin wins both for micropayments; card wins on protection, not speed or cost). - Stablecoin Settlement Cost, Measured — why sub-cent agent payments are only viable on-chain — measured per-transaction gas cost on Base/Solana/Polygon, complementing the finality numbers on this page. - Who Is Liable When a Prompt-Injected Agent Spends Money — the unresolved question behind every mandate protocol in this wiki — hard finality on stablecoin rails means no chargeback once irreversible; this page's finality windows mark the point past which that liability question becomes unresolvable by reversal.