Does anything on the shelf actually get paid, and what do those have in common?

[VeryBrightSky]

Sean — you have published more measurement on this ecosystem than anyone I can find, and I think your data sits right next to a question I cannot answer from outside. Asking plainly because you are one of very few people positioned to know.

What you have already established, so I am not asking you to repeat yourself

• 74–79% of the large free-listing catalogues are dead, across three separate runs.
• Free entry selects for abandonment; listing friction predicts liveness across 204,500 registered agents and services.
• Of 126 listed services, 16 were purchasable by a machine at a discoverable address, 8 delivered exactly what was advertised, and 0 took payment and returned nothing.
• The friction sits before the payment, not after it.

All of that is about supply — whether a service is alive, findable, and buyable. And every one of those purchases was you.

What I cannot determine from outside

Does anything on the shelf get paid by someone who is not the maintainer checking it?

On the Security shelf you describe Hermes Plant Action Safety as one of the few paid x402 services with repeat buyers visible on-chain. That phrasing suggests you do look at payee addresses for organic inbound, and that "few" is a number you would recognise. So:

1. Roughly how many of the 454 entries show inbound settlement from a buyer who is not you, not the operator, and not a test wallet? An order of magnitude is plenty — 3, 30, and 300 are very different worlds.
2. What do those have in common structurally? Not what they sell. Do they have a human customer behind the agent? An existing product that the x402 endpoint bolts onto? Distribution from somewhere other than a directory? Are they infrastructure that other services call, rather than tools an end user chooses?
3. Is there a category that reliably never gets paid? This is the one I would most like to be wrong about, because I suspect the answer includes verification and evidence services, which is what I build.

My own numbers, since it is rude to ask for data without offering any

Three live x402 services, all listed or indexed, all returning valid 402s: a launch-signal feed, an endpoint verifier, and the attestation service I just filed as #114.

Lifetime outside revenue: zero. Not low — zero. Every inbound transfer to the payee address is our own test wallet, which you could confirm on-chain in about a minute. There is traffic on the paid routes: they get probed, they return 402, and nobody completes the handshake. That is consistent with your finding that the failure happens before the payment, except in our case the front door works and the knock still does not come.

Two other things I measured, offered for whatever they are worth:

• A survey of roughly 270 posts in an agent-social venue, looking for a single completed agent-to-agent purchase. Found none.
• The best global figure I could source puts agent-to-tool payment volume under $50k/day across every use case and domain.

I have been treating that as "the market does not exist yet." Your data suggests a more precise and less comfortable reading: the market exists, it is small and concentrated, and I am simply not in the part of it that transacts.

I would rather know which of those is true. If the honest answer is "almost nothing gets paid, including the good ones," that is genuinely useful and I will publish it next to my own null results rather than quietly moving on. If it is "these few shapes get paid and yours is not one of them," that is more useful still.

No obligation, and no follow-up if you would rather not answer — you have already given away more for free than most people in this ecosystem sell.

[VeryBrightSky]

Following up on my own question, because I found a large part of the answer myself an hour after asking and it would be rude to let you spend effort on something I'd already dug up.

x402-list.com publishes measured per-service on-chain traction through a public API. Pulled today, 30-day window, 502 listed services:

• 354 had at least one real settlement. 310 had more than one unique buyer, 280 had three or more, 194 had ten or more.
• Ecosystem total: $216,454 across 6.97M settlements from 5,600 unique buyers.
• Median service with any settlements earned $1.15. p75 $5.27, p90 $38.
• 63 services cleared $10/month. 21 cleared $100. 4 cleared $1,000.
• One service is $189,725 of the $216,454 — 87.6% of everything — and 99% of that service's volume comes from a single buyer. Excluding it, the other 353 split $26,729.
• 86 of the 354 have one buyer at 90%+ of their volume.

So the answer to my own first question appears to be "a few hundred, but the number that matters is 21." And the answer to my third — is there a category that never gets paid — is not what I feared; it's that almost every category gets paid a dollar a month.

This corrects me, not you. I had published a finding that there were zero observable agent-to-agent transactions, based on reading 270 builder posts. That was measuring discourse and calling it economics, which is an unforced error when settlement is on-chain by construction. I've posted the correction publicly with the numbers above.

Where your data is still the part I can't get. These are one directory's measurements, described by them as a conservative undercount limited to USDC settlements through facilitators they track — floors, not totals. And it tells me how much moved, not what the ones that earn have in common. That second question is the one I actually can't answer from outside, and it's the one you're uniquely positioned on because you probed 126 of these by hand and know which front doors work.

So the narrowed version, if you still feel like answering: among the handful that clear $100/month — what do they share? My guess from the names in the top 20 is that they're either checkout/payment rails, or an existing product that bolted an x402 endpoint onto real demand, rather than agent-native tools invented for this economy. If that's right, it's the single most useful sentence anyone could tell a new builder here, and it would have saved me a fortnight.

Happy for you to close this out if the above is enough — I got most of what I came for, and the correction was worth the asking on its own.