Facilitators probed

Inclusion is a property of the facilitator’s own conduct rather than an opinion TOLL holds about it. Serving the extension openly is a thing a facilitator does or does not do, and the request that establishes it is the one printed below, which anyone can repeat and check against this page. Everywhere else this site refuses to make quality judgements about what it measures; deciding which catalogues to read on the basis of reputation would be the same judgement, moved one step earlier and made invisible.

What the rule costs, stated so it is not discovered later:

• It does not permit leaving out a catalogue for being concentrated in one seller.
• It does not permit dropping entries that look like noise; they are counted and described.
• It admits facilitators TOLL knows nothing else about, which is the point: knowing something else about them would be the judgement this rule exists to avoid.

One request per facilitator, a stated user agent, a stated timeout, taken once on 2026-09-15 09:44 UTC. The site never probes on a page render.

curl -sS -A 'TOLL-index/0.1 (+https://tollindex.com; coverage survey)' \
  --max-time 15 \
  '<facilitator base>/discovery/resources?limit=2'

PayAI lists 28,662 resources, which is more than this index holds in total. That number and the number of endpoints a reader could call are not the same number, and the difference is not small. A sample of 800 resources, taken at evenly spread offsets rather than from the head of the catalogue, found 10% whose resource field is not a fetchable URL at all — entries under schemes like monopoly:// with no host — and 50% belonging to a single seller, orbisapi.com. The sample spans 115 hosts in all.

So “28,662 resources” and “28,662 addressable endpoints” are different claims, and only the first is supported. TOLL is not saying the catalogue is wrong: enumerating many paths under one host is a legitimate thing for a seller to do, and a scheme TOLL cannot fetch may be perfectly meaningful to whoever wrote it. It is saying that a headline count carries a shape, and that the shape has to be published with it. The exact figure is unknown until the catalogue is pulled in full, which has not been done.

Of the sample, 38 resources (4.8%) are already in this index by canonical URL, and 14 of its 83 addressable hosts are hosts TOLL already holds. The overlap is small: this would be mostly addition rather than duplication.

Every usage figure on this site — calls, unique payers, calls per payer, the concentration ranking, the correlation, the density field, a category position — rests on a quality block carrying a call count, a unique payer count and a last-called time. That block is a Coinbase addition. It is not in the bazaar discovery extension, and of the 4 facilitators serving discovery openly, 1 carries it: Coinbase. The other 3 — PayAI, Thirdweb, Dexter — publish the listing and nothing about its use.

Today 52,065 of 52,265 listed endpoints carry counters, 99.6% of the index. Reading the other three catalogues would add 32,703 resources gross, or about 31,150 after the overlap the sample measured, and the counters would then cover roughly 62% of what the index holds. Nothing about the measurements would change; what would change is what they are measurements of. “Most called” would mean most called among a third of the listings, and every page stating a usage figure would have to say so on the same line.

That is a product question rather than a data one, and it is not decided here. This page exists so the question is visible before an ingestion plan is written rather than after. Nothing from these catalogues has been ingested.

Two limits on this page itself. The directory it starts from is a view of the ecosystem and not a census: Binance B402, the index’s second source, is not listed on it at all, so a rule that read only what the directory carries would inherit the directory’s omissions. And a facilitator is not the only way an endpoint reaches the world — a seller who publishes a manifest at /.well-known/x402 on their own domain appears in no catalogue, and no amount of probing facilitators would find them. The first limit is a decision about sources. The second is a shape this method cannot reach, and it is stated on the method page for that reason.