What x402 on Arc actually is

Basis. All classified figures come from classification run 6, a scheduled run after the W1 fix, as of 2026-09-24 04:50:26.064 UTC, methodology v1.0 (final), rules hash 0eb0d034931a. Scope unless stated: USDC EIP-3009 transfers in the public phase (from block 21,056,228, 2026-09-15 22:15 UTC) up to the as-of time. One transfer landed at the as-of second, after the run started, and is excluded, leaving 74,741 transfers, 1,735,222.39 USDC. Every figure is given by count and by value, because different things dominate each.

1. Headline

Of 74,741 settlements (1,735,222.39 USDC) made through EIP-3009 on Arc since the public launch, 250 settlements (0.33%) worth 41.23 USDC (0.0024%) look like agent commerce.

What fills the gap depends on how you measure:

• By count, one wash ring dominates: 61,147 of the 61,198 wash settlements come from a single relayer's 53-wallet ring (§4).
• By value, one single-payer settlement pipeline dominates: 1.60M USDC, 92.0% of all value, where no classification rule fires (§5). It sits in unknown, which holds 99.66% of value.

2. Class breakdown

Public phase, run 6. All seven classes are shown, including those with no settlements. Unknown is its own category: nothing is folded into it or out of it.

• Unknown means no v1 rule matched: not submitted by a labeled facilitator, and no test, funding, refund, wash or script signal. It is not a residual of the other classes, and it is not commerce. 0x7347's pipeline (§5) alone is 1,895 of these settlements and 1,596,618.01 USDC.
• Script: 0. No payer→payee pair shows clockwork timing (S1), and no single-target payer uses identical amounts (S2).
• Refund: 0. No labeled service has paid a refund on-chain (§8).
• Private phase, for reference: 201 transfers, 11,453.34 USDC (200 unknown, 1 test).

3. Who settles it

Settlements by a labeled facilitator (x402-likely): 312 (0.42% of count), 43.27 USDC (0.0025% of value). 250 of them classify as Likely agent. By methodology, settlements by unattributed relayers can't be Likely agent, even if some are. Attribution coverage is the main limit on the agent figure.

Largest Likely-agent payee: ArcRouter (arcrouter.co), 9 settlements, 27.00 USDC, all through Arcus.

4. Case: the busiest relayer is a closed wash ring (0x2b9a…055c)

0x2b9a3197ed35d56e2e1c2a01f4d649586821055c, unlabeled. Public phase, run 6:

• 88.0% of all public settlements by count, 0.0074% by value.
• 56 payers and 55 payees, 53 of them on both sides. 5,675 distinct amounts, mean 0.00196 USDC. Payments bounce back between the same wallets within minutes. Active 2026-09-17 07:43 → the as-of time.
• Classified by v1 round-trip rules (W2/W3). The 4,566 unknown settlements evade W2's ±5% amount tolerance through randomized amounts. Treatment in v1.1 (a closed-ring rule) is under consideration.
• Wicket (wicketworld.xyz) ranks this relayer #1 by count (59,354 settles as of 2026-09-23 18:10 UTC).

5. Case: one payer, one contract, most of the value (0x7347…)

Relayer 0x734713ed06d87b623c11ded486cfcec5728cd691 submits settlements from a single payer into a single contract (0xe742f9df…, which is also the payee). 1,895 settlements, 1,596,618.01 USDC, 2026-09-16 15:12 → 2026-09-24 04:44 UTC, with 1,864 distinct amounts.

• 92.0% of all public-phase value (2.5% of count).
• No classification rule fires on any of its settlements. It is not a facilitator Farelight can attribute; there is one payer, so no wash cluster or round trip; and amounts and timing are too irregular for the script rules. It stays in unknown.
• The transfers go through a contract (tx.to is the contract, not USDC), so indexes that only count direct USDC calls, such as Wicket, don't include it. That's consistent with our reconciliation (§7).
• A single entity settling with itself through EIP-3009 is not evidence of agent commerce, whatever it is.

6. Case: most of the rest of the value is funding (0x1fdc…cf93)

0x1fdc02956b5bfc09780b3d4b492d4bc0906ecf93, unlabeled. Public phase, run 6:

• 5,000.00 USDC is one transfer, 0x80d31692… → 0xf48a7176… (tx 0x759f46f5…9d6e, 2026-09-16 18:21 UTC), classified funding (F1a): it is the recipient's first funding, and the recipient then spent.
• The recipient then sent 1.00 USDC to 646 new wallets, a fan-out classified funding.
• Wicket ranks this relayer at 97.2% of the value it indexes (6,951.02 USDC as of 2026-09-23 18:10 UTC). Under Farelight's methodology, 81.2% of its value is funding. The payment-sized remainder (0.01–0.05 USDC) is under ~10 USDC.

7. Cross-checks: independent methods agree

• Wicket reconciliation (run 2026-09-24 02:50 UTC, complete backfill). On all 20 fully covered days, 219 of 219 day × submitter rows match exactly: 65,189 settlements and 7,163.359843 USDC on both sides, under Wicket's inclusion rule (from block 21,108,966, tx.to == USDC). Our extra volume outside that rule is itemized: private phase, Gateway deposits, and contract-routed settlements such as §5.
• Arcus. The operator claims 117 settlements / $35.11. A chain-state reconstruction of its published signer found 96 / 35.207908 USDC on mainnet (every tx a settlement; +24 / 3.50 on testnet), and Wicket independently showed 96 / 35.21. The index matched exactly: 96 at the audit cutoff, 102 / 37.708908 on 2026-09-24, equal to a fresh chain audit. As of run 6: 104 / 38.708908. The operator's count claim doesn't reconcile; its volume is within 0.3%.
• Arclight. 9 settlements / 0.0085 USDC, matching its signer's nonce.
• Circle. Its /supported publishes no signers. Attribution rests on Farelight's own observed settlement (0xa3df…6f5f).

8. What a real service looks like (ArcRouter)

• arcrouter.co's treasury (0x6bf0…e9a9): 9 settlements, 27.00 USDC, all Likely agent, all via Arcus.
• It charges a ceiling up front and says it refunds the unused part. No refund has been paid on-chain: the treasury balance reconciles exactly to inflow − buyback − gas.
• One buyback-and-burn: 8.48 USDC → 380,741 $ROUTER → 0x0 on 2026-09-21. That's 79.5% of its claimed 30-day volume; the "80% buys and burns" claim has not been verified over time.

9. Beyond x402 settlements

• Gateway deposits are the only on-chain trace of Circle's off-chain nanopayments. They are not counted as settlements above. Public phase: 10 deposits, 28.00 USDC (4 payers). Private phase: 1,355 deposits, 847,084.05 USDC (1,088 payers), mostly through one relayer.

10. Method and limits

Full methodology: /methodology (v1.0, final).

• Precedence: test → funding → refund → wash → script → agent → unknown. Only agent counts as commerce, and it is labeled Likely agent because on-chain data can't prove an AI made the payment.
• A payment is never a funding edge. A service's first inbound USDC is usually its first customer's payment; that's a sale, not common control (2026-09-24 fix).
• Phase boundary: block 21,056,228 (2026-09-15 22:15 UTC), derived from sender-growth on-chain, approximate to minutes.
• Settlements by unlabeled relayers can't be Likely agent, and refunds paid by plain transfers aren't yet read. Both are coverage limits, stated as such.