SECOND MEASUREMENT SM-006
x402 payments on Base fell by three-quarters in two months, and the same two operators still send most of them
x402 lets programs pay each other pennies over the web, and a July study found most of the 136.7 million payments on Base were operators paying themselves, and at least a fifth were manufactured outright. We counted one day in August: 234,490. We counted again on 1 October: 63,003, three-quarters gone. The busiest payers changed; the one behind 56.9% of August's traffic made none. The senders didn't: the same 20 wallets in two groups sent about 80% of the payments both days.
- A July 2026 study counted 136.7 million x402 payments on Base and found most were operators paying themselves, at least a fifth manufactured.
- Our one-day count for 5 August, 234,490, reproduces exactly.
- On 1 October: 63,003, about three-quarters fewer.
- The payer behind 56.9% of August’s payments made none on 1 October.
- The same 20 wallets in two groups sent about 80% of payments on both days.
Disclosure: Markovian Protocol holds no financial position in any organisation named here, was paid by no one for this work, and showed it to no one before publication except the organisation measured. How we work.
An x402 payment shows up on Base as a USDC transfer signed by the payer and sent to the chain by someone else, who pays the fee. That someone is the facilitator. Every such transfer leaves a marker, an AuthorizationUsed event, so counting markers counts payments without anyone's database in between. We count every marker. The study counts only markers sent by 129 wallets on its list of known x402 facilitators, so our number is an upper bound on theirs. On the study's own data the two differ by 2.5%.1
What they said, what we found
| The study said | We found |
|---|---|
| 136,708,672 x402 payments on Base over 280 days to 23 June 2026, worth $44.1 million1 | Our August count reruns exactly: 234,490 in the day ending 6 August 03:47 UTC, the same to the payment. |
| From January to June 2026 the pace was “about 145,500 settlements per day”1 | 234,490 on 5 August, then 63,003 on the day ending 1 October. That is 43% of the study's January–June pace. |
| A single dominant facilitator on each chain1 | Two groups of wallets that move in step sent 80.4% of payments in August and 80.0% in October. |
| Much of the traffic is manufactured: “at least 21.20% in count” is fictitious1 | We didn't test this. What changed is the shape: in August one payer made 56.9% of all payments; in October the busiest made 10.3%. |
Rerunning August
We published a one-day count in August with the code to repeat it.2 Rerun today on the same blocks, it gives the same answer: 234,490 payments from 3,704 payers, 200,530 of them matched to an amount totalling $558,695.15, and the same top 15 payers in the same order.
Rerunning it also turned up a mistake in our code. Five lines in the August write-up describe the busiest payers: that 34 of the top 40 had never sent a transaction, how many wallets submitted for the busiest one, that no USDC came in, and their balances. Our script printed those lines as fixed text instead of measuring them, so anyone rerunning it on a different day would have been shown August's facts. We've fixed it and measured them. For August they hold: 34 of 40, about 20 submitters, no USDC in. The balances come out a little different ($6,991.79 for the busiest, against $6,892.99 published), because we'd read them at some later moment we didn't record. They are now read at the last block of the day being measured.
Two months later
| 5 August | 1 October | |
|---|---|---|
| Payments | 234,490 | 63,003 |
| Different payers | 3,704 | 8,478 |
| Busiest payer's share | 56.9% | 10.3% |
| Top 40 payers' share | 77.6% | 51.3% |
| Busiest payer→recipient pair | 66.6% | 11.2% |
| Value matched to a payment | $558,695 | $452,910 |
| Median payment | $0.015 | $0.005 |
| Top 40 payers that never sent a transaction | 34 | 23 |
One day each: Base blocks 49,556,153–49,599,353 (2026-08-05 03:47 to 2026-08-06 03:47 UTC) and 52,006,727–52,049,927 (to 2026-10-01).
Fewer payments, but more than twice as many payers, and none of them dominant. The one payer that made more than half of August's traffic is gone from the top. The value moved fell less than the count did, so the average payment got bigger even as the typical one got smaller.
Five of October's top 40 payers are ordinary wallets that have borrowed contract code under EIP-7702, a feature that lets a wallet act like a program. The busiest payer is one of them. None of August's were.
The two groups of wallets
The wallet that sends a payment to the chain pays its fee, and each wallet's lifetime transaction count goes up by one every time. Wallets run by different people drift apart. Wallets run by one system stay together.
| Group | Wallets | Lifetime transactions each | August share | October share |
|---|---|---|---|---|
| A | 15 | about 1.57 million, all within 0.37% | 61.1% | 60.1% |
| B | 5 | about 3.33 million, all within 0.38% | 19.3% | 20.0% |
Shares are of payments in 400 randomly chosen blocks per day: 2,213 payments in August, 837 in October. Transaction counts read on 2026-10-01. Both groups are the same addresses on both days: 15 of 15 and 5 of 5.
The traffic fell by three-quarters and its busiest payers changed: the one behind 56.9% of August’s traffic made none, and only 7 of August’s top 40 are still in October’s, but the same two groups of wallets still send about four in five payments.
The evidence
Saved 2026-10-02T18:12:56Z. Copies and SHA-256 hashes: SHA256SUMS. Every address can be looked up on any Base block explorer.
Exhibit 1 · The study’s count
“Our rule is the intersection of the event with a curated facilitator allowlist, which tightens the universe to the 136,708,672 x402 settlements ($44,121,383.81).”
arXiv:2607.12575 · saved copy x402-study-arxiv-2607.12575.pdf
Exhibit 2 · The study’s pace for early 2026
“the January–June 2026 tail runs at about 145,500 settlements per day”
same copy
Exhibit 3 · Group A, the same wallets both days
0x14fda13953fc30428938e6bf950d036e77214e52 0x2a89407a98a0732b7fd578c4e156b7166540eb5a 0x42dd53906b49c202e8e934b059dc019e04634b00 0x4c934c63c786157fefd990945b25ea60a0fb0205 0x59b7ebc67a3d627fabaf06768c818638452ae704 … 15 wallets in all 5 August: 61.1% of sampled payments 1 October: 60.1% of sampled payments Same 15 addresses: yes
our runs · raw fleets-2026-08-05.json, fleets-2026-10-01.json; block windows in window-2026-08-05.json, window-2026-10-01.json
Disclosure timeline
| 6 Aug | First one-day count published on GitHub. |
| 1 Oct | Recount and replay; published; sent to the study’s authors. |
Waiting on the study’s authors since 1 Oct.
What we can't be sure of
- We count every EIP-3009 transfer, not just those from listed x402 facilitators, so both days are upper bounds. On the study's data that difference was 2.5%; we haven't measured it for these days.
- Two single days, not a trend. A busy or quiet day would move these numbers.
- The group shares come from about 1% of each day's blocks. October's sample is small (837 payments), so its shares are rough.
- Two groups moving in step shows each is run by one system. It doesn't show who runs them, or that anything is wrong. A facilitator running a pool of wallets is normal.
- We didn't test the study's main finding, that much of the traffic is manufactured.
Run it yourself
The scripts are on GitHub: github.com/MarkovianProtocol/x402-second-measurement. They use only Python's standard library and Base's public endpoint.
X402_DIR=./aug python3 x402_pull.py - 49556153 49599353 # the August day X402_DIR=./oct python3 x402_pull.py - 52006727 52049927 # the October day # then, in each folder: python3 x402_values.py # amounts python3 x402_payers.py # the busiest payers python3 analyze.py # findings.md python3 x402_facilitators.py 400 && python3 x402_fleet.py # the wallet groups
Reproduced from the public repository on 3 October 2026 on a clean machine: exact (234,490 and 63,003 settlements; the same 15 and 5 wallets at about 80%). The per-payment value scan runs longer than ten minutes. Full table on the track record.
What you can do with this
- If you quote x402 volume, use the two-day counts and the fleet share, not the headline total; most settlements come from twenty wallets.
- If you build on x402, measure your own counterparties' nonce and contract status; the busiest payers are fresh accounts.
References
- Shengchen Ling, Yajin Zhou, Lei Wu, Cong Wang. How Agentic Is Agentic Commerce? A Population-Scale Measurement of x402 Adoption and Authenticity. arXiv:2607.12575, July 2026. arxiv.org/abs/2607.12575. Sections 3.1 and 5.5.
- Markovian Protocol. x402 second measurement. 2026-08-05. github.com/MarkovianProtocol/x402-second-measurement.
Cite as
@misc{markovian-sm006,
author = {{Markovian Protocol}},
title = {x402 settlement on Base, two days two months apart},
number = {SM-006},
doi = {10.5281/zenodo.23122988},
year = {2026},
month = oct,
url = {https://markovianprotocol.com/measurements/sm-006.html}
}