x402 DACH Radar, Issue 2 (August 2026)
x402 DACH Radar, Issue 2. Primary measurement from 13 paid API endpoints we operate ourselves on Base mainnet. Measurement window 1 to 31 August 2026 — the first full calendar month; Issue 1 was a three-day pilot and is not comparable to this issue in absolute terms.
Every figure marked “own measurement” comes from our own demand counter. Every factual statement carries a confidence tag, every headline claim carries a falsifier. What we did not measure is labelled as not measured.
Published by OzDreamWalk. Not legal and not tax advice.
Part A: Report
1. The short version
- Price-request traffic exploded — real demand did not. 68,280 price requests in August against 517 in the three-day pilot. Strip out the empty probing envelopes (Section 4) and 3,832 real, content-bearing price requests remain — fewer per day than in the pilot (124 versus 194 per day). The jump is mostly a measurement artefact of our own decision on 27 July, not demand growth.
[HIGH, own measurement] - Two foreign payments, both new, both on the same route, the same identifier. 0.01 USDC on 8 August from Singapore, 0.01 USDC on 30 August from the US — both on
agent-ready, both under the identifiernode, both from different, on-chain distinct addresses. Cumulatively since measurement began (23 July) that makes four independent external payers, zero repeat purchases among all four.[HIGH, own measurement and chain proof] - Conversion stays in the per-mille range. 2 of 68,280 price requests were paid, that is 0.003 percent. Against real (non-empty) price requests: 2 of 3,832, or 0.052 percent — half the 0.10 percent from an internal follow-up measurement in the five-day window right after Issue 1.
[HIGH] - Not a single self-test purchase all month. Our own test buyer wallet stayed empty (a measurement question left open since Issue 1, carried forward verbatim). For a full month we produced no end-to-end confirmation of our own that the routes are actually payable — only foreign traffic and two foreign payments still prove that.
[HIGH, own measurement] - The directory has stopped shrinking — but almost all our own entries are gone anyway. Third data point: 14,681 entries (1 September), after 24,910 (23 July) and 14,376 (26 July) — size has stabilised. Yet of our 13 routes, only one is still listed in the directory (
agent-ready), although six were still listed in July and none of them have been renamed since. Details in Section 6.[HIGH/MED, see there] - First server errors since we started measuring. Two 5xx responses for the whole month, both triggered by known verification services, no refunds needed.
[HIGH]
2. How this was measured
Same construction as Issue 1, now across 13 rather than nine paid routes (added since Issue 1: iban, vat-format, robots-lint, x402-demand). Every call leaves one row: timestamp, route, outcome, truncated identifier, country code, price, payer address, settlement hash, raw HTTP status. No IP address, no request body.
Since this issue, our own traffic is excluded via a classification rule kept in code, no longer via the raw SQL pattern list from Issue 1 — an internally documented change from 26 July. It changes nothing about this issue’s numbers (same ozdw-/hans convention, same payer address list), but it makes the report and the paid x402-demand endpoint draw from the same source of truth. 31 of the 90,712 rows in the window are our own (all free, catalog self-checks) — markedly fewer than the 329 of 7,660 in the pilot, because not a single self-test purchase ran in August (see Section 8).
3. Who is knocking
248 distinct identifiers (plus one group with no identifier, 1,233 rows), 90,681 foreign requests. The ten strongest:
| Identifier | Requests | Routes | Price requests |
|---|---|---|---|
mako-pulse-prober/0.1 | 19,023 | 3 | 19,022 |
CarbonMonitor/0.1 healthcheck (+https://carbon-cashmere.de) | 15,984 | 13 | 15,984 |
x402-observer/1.0 (uptime+trust monitor; +https://x402.fuchss.app/trust) | 9,342 | 13 | 9,341 |
forum-labs-trust-prober/1.0 (x402 endpoint QoS monitor; …) | 7,564 | 13 | 7,564 |
node | 5,213 | 13 | 3,099 |
AgentReeve/5.1 (+https://agentreeve.com) | 3,411 | 13 | 156 |
MPP32-Health/1.0 (+https://mpp32.org) | 3,014 | 13 | 1,507 |
entropy-daemon-trust-oracle/2.0 | 1,677 | 13 | 0 |
CoinbaseBazaarDiscovery/1.0 (+https://docs.cdp.coinbase.com/x402) | 1,602 | 13 | 1,602 |
stelar-trust-monitor/1.0 | 1,508 | 13 | 0 |
All ten query all or nearly all 13 routes — a catalog walk, not interest in a specific answer, exactly as in Issue 1. The full list (248 identifiers) is not published this issue — too large for a readable appendix, and several identifiers carry an email address in their free-text tail. Only identifiers that are unambiguously a self-declared public service appear here, truncated.
Origin by country (40 countries total, none without a country code): US 53,340, Germany 28,138, France 3,447, Netherlands 1,634, Japan 1,166, Switzerland 940, Denmark 719, Belgium 602, Finland 356, Singapore 102, Canada 70, all other 29 countries under 30 requests each (sum 231). [HIGH]
New compared with Issue 1: not a single row without a country code. In the pilot it was missing on four of 7,331 rows; in August, on none of the 90,681. No known reason, just recorded. [MED]
4. What actually saw a price — and the biggest delta versus Issue 1
68,280 price requests — but 64,448 of them (94.4 percent) are empty envelopes, meaning a call with {} or no body at all. Only 3,832 price requests carry actual content. [HIGH]
This is not a market movement, it is the direct consequence of our own operational decision on 27 July: since then, an empty body and an empty {} envelope get a 402 instead of a free no — the reason was that a foreign trust index otherwise reported “not payable” even though the route is payable. The pilot figure (517 price requests, 31.9 percent of them empty per the same follow-up measurement) predates this rule. A large part of the apparent demand growth is therefore the same probing traffic now counted differently, not more interest. [HIGH, cause documented and verified]
Counting only real price requests, demand actually falls per day: 3,832 over 31 days is 124 per day, against 964 over 4.98 days (193.6 per day) in the five-day window after Issue 1. This is the less flattering number, and the more honest one. [MED, comparison of two different window sizes — the original window was a pilot follow-up, not a calendar month]
Price requests per route (all 402s, including empty envelopes — a route-by-route split of the empty share was not separately counted this issue):
agent-ready 12,006 · deadlines 9,940 · bfsg 9,886 · holidays 3,995 · llmstxt 3,672 · readability 3,663 · imprint 3,620 · iban 3,618 · invoice 3,612 · outreach 3,588 · vat-format 3,587 · robots-lint 3,561 · x402-demand 3,532. Sum 68,280 (reconciliation identity 4, Part B).
Catalog (free endpoint) was fetched 102 times.
5. The August payments
Two foreign addresses paid in August, both on agent-ready, both 0.01 USDC, both under the identifier node:
| Time (UTC) | Payer | Country | Identifier | Tx |
|---|---|---|---|---|
| 2026-08-08T22:32:39Z | 0x1392…3d7aD | SG | node | 0x55a22565…0331 |
| 2026-08-30T15:27:39Z | 0xC533…92D7A | US | node | 0x322aadec…8a971 |
Both block-pinned and verified (settlement hash present, outcome=paid), both from addresses that are neither our own test buyer nor either July address (Dexter-Verifier 0x7e571E95…, payments-mcp-proxy 0xfe2d5E…). Cumulatively since measurement began, that is four independent external payers, zero repeat purchases among all four. [HIGH, own measurement and chain proof]
Same identifier, two provably different actors. Both payments carry the identifier node — the generic default identifier of the Node.js runtime, not a distinct piece of software. Without the payer addresses these would be two indistinguishable rows; with them, they are two different on-chain addresses. That is a concrete piece of evidence for the standing doctrine “identifiers are not actors” — here in the less common direction: not one actor under many identifiers, but one identifier under several actors. [HIGH]
What this does not prove: that an agent needed the assessment for its content. node as an identifier reveals nothing about the purpose of the call — it could be a script, an agent framework with default headers, or once again a verification service. Both callers made exactly one request each, on agent-ready only, none on any other route; that pattern differs from the catalog-walking identifiers in Section 3.
6. The directory rule — a third data point, and an extension
The third directory data point is in, and it answers the open question from Issue 1: the directory has stopped shrinking. Paged exhaustively on 2026-09-01: 14,681 entries, against 24,910 (23 July) and 14,376 (26 July). Size has roughly stabilised since late July (+2.1 percent versus 26 July, −41.1 percent versus 23 July). [HIGH, own measurement, third independent survey]
But: of our 13 routes, only a single one is still listed in the directory. agent-ready — with lastUpdated 2026-08-30T15:27:39.854Z, exactly 544 milliseconds after the second August payment (15:27:39.310Z). That confirms again the rule known from Issue 1: the entry hangs on the last settlement. [HIGH]
New, and not observable in Issue 1: six of our routes were still listed on 26 July (agent-ready, bfsg, holidays, invoice, llmstxt, outreach, plus the test entry deadlines) — five weeks later, all but agent-ready are gone again, without any of these URLs having been renamed since. The rule known from Issue 1 (“renaming drops you out”) does not explain that; it only covers the case of a path change. The natural extension: an entry also decays without a rename, if it goes unpaid long enough. [MED, one observation, threshold unknown] Falsifier: an entry that has gone more than five weeks without a new payment stays listed anyway — that would refute the decay theory and the drop would need a different cause. This issue cannot distinguish the two, because we have not observed a second unpaid, unchanged entry over a known period.
7. Market context
No new external study was researched for this issue — the figures cited in Issue 1 (median lifetime revenue $3.96 per resource, ~52 percent reachable hosts, a highly concentrated facilitator landscape) are a month old and are not repeated here as current claims, only referenced. An updated external survey is pencilled in for Issue 3, provided a new primary source with a documented method can be found.
8. What operations showed in August
Not a single self-test purchase all month. Our own test buyer wallet, per the measurement question left open since Issue 1, was already empty by late July; it was not topped up in August. Result: 0 self paid rows (Issue 1 had 20 in three days). Without our own test runs there is no regular self-confirmation that every route is still actually payable — the only confirmation this issue are the two foreign payments and the passive observation that price requests keep getting a 402 response. Top up before the next issue, or operational confirmation stays accidental instead of planned.
First server errors since measurement began, both from known verification services. Two 5xx responses across the whole of August: bfsg on 16 August (caller x402-observer), agent-ready on 29 August (caller mako-pulse-prober). No refund situation, because both failed before the payment gate (no tx). Issue 1 had zero 5xx cases in the pilot; whether this is a real fault or ordinary noise at this volume cannot be decided from two cases out of 90,681 requests.
Correction (2 September 2026): this section originally stated something false. The first version said x402-observer, the third-strongest identifier in the entire August dataset with 9,342 requests, was internally classified as unclassifiable rather than as a verification service. Tested live against our own classifier instead of citing the underlying internal note: that is not true, the identifier has always been correctly classified as a verification service. The internal note we had cited was itself wrong and has since been corrected. No rule change is needed, and none of this issue’s nine metrics are affected.
9. What this means for a data provider
Price-request growth is not a demand signal unless you know how the counting rule itself changed. The 132-fold rise in price requests versus the pilot (517 to 68,280) looks dramatic and is mostly our own decision, not an outside movement: a spec-compliance fix has, since 27 July, turned an empty test call into a 402 response instead of a free no. Anyone comparing a metric over time has to carry every rule change of their own along with it, or they are comparing two different measurements under the same name.
Strip the rule out, and actual demand in August is as small as in the pilot, if not smaller. 124 real price requests per day against 194 in the pilot. Two paying addresses in an entire month, both new, not a single return visit. Anyone building a revenue model on this portfolio should plan with numbers at this scale, not with the eye-catching 402 curve.
Discoverability stays fragile, and now also transient. Issue 1 showed: the entry hangs on the last settlement. Issue 2 adds: it apparently does not hold indefinitely without a new one either. A data provider that does not settle regularly — even if only through genuine foreign traffic or its own test purchases — disappears from discoverability, entirely without acting.
10. What this does not prove
| Claim | What would overturn it |
|---|---|
The price-request jump is mostly a measurement artefact of the 27 July {} rule | A count of price requests with content also shows a multiple versus the pilot |
| Real daily demand in August is no higher than in the pilot | A repeat measurement of the five-day window after Issue 1, with the identical rule, yields a markedly lower daily figure than reported here |
| Four independent external payers since measurement began, zero repeat purchases | One of the four addresses appears a second time as a payer on any route |
| Directory entries decay even without a rename | An unchanged route that has gone unpaid for more than five weeks stays listed |
| The directory has stabilised at roughly 14,000–15,000 entries | A fourth measurement lands clearly outside this range |
One limit carries over unchanged from Issue 1: 248 identifiers are not 248 actors, and we cannot resolve that — Section 5 even shows the reverse case, two actors under one identifier.
11. Next issue
Carried forward verbatim until answered: (a) the observer classification decision (an internal rule proposal exists), (b) whether the test buyer wallet has been topped up and self-tests are running again, (c) whether any of the four known payer addresses returns, (d) whether agent-ready, as the last remaining directory entry, also disappears if no new payment lands in September — that would be a further data point for the decay theory in Section 6.
Method: this rests on our own demand counter across 13 self-operated paid endpoints, window and ceiling in the header of this document. The underlying queries are frozen so future issues stay comparable. The biggest interpretive conflict in this issue: the price-request increase is mostly, but not entirely, attributable to our own rule change — how much of it is additionally real growth cannot be separated from the available data, because the rule and the month changed at the same time.
Part B: Data Appendix
B1 Reconciliation identities (pin id <= 115789, window 2026-08-01 to 2026-08-31)
| # | Identity | Result |
|---|---|---|
| 1 | total = own + foreign | 90,712 = 31 + 90,681 ✓ |
| 2 | foreign = sum of outcomes | 68,280 + 22,295 + 102 + 2 + 2 = 90,681 ✓ |
| 3 | sum of countries (40) = foreign rows | 90,681 = 90,681 ✓ (0 rows without a country code) |
| 4 | sum of 402s per route = 402 total | 68,280 ✓ |
| 5 | settlements with a hash | 0 own + 2 foreign = 2 ✓ |
B2 Foreign outcomes, interpreted against raw
| Outcome (interpreted) | Count | Raw status | Count |
|---|---|---|---|
402 | 68,280 | — (no separate raw status recorded for 402) | — |
400 | 22,295 | 405 (wrong method) | 22,071 |
| 400 (rejected body) | 224 | ||
free | 102 | 200 | 102 |
paid | 2 | 200 | 2 (both carry a hash) |
500 | 2 | 500 | 2 |
Of the 68,280 price requests, 64,448 (94.4%) carry an empty body or envelope; only 3,832 (5.6%) are real, content-bearing price requests.
B3 Price requests per route (foreign, incl. empty envelopes)
agent-ready 12,006 · deadlines 9,940 · bfsg 9,886 · holidays 3,995 · llmstxt 3,672 · readability 3,663 · imprint 3,620 · iban 3,618 · invoice 3,612 · outreach 3,588 · vat-format 3,587 · robots-lint 3,561 · x402-demand 3,532.
B4 Foreign identifiers — top 10 (of 248 + 1 without an identifier)
See Part A, Section 3. The full list is not published (size, some identifiers carry email addresses in their free-text tail).
B5 Countries (foreign, 40 total)
US 53,340 · DE 28,138 · FR 3,447 · NL 1,634 · JP 1,166 · CH 940 · DK 719 · BE 602 · FI 356 · SG 102 · CA 70 · ID 25 · GE 24 · LT 14 · IN 13 · BR 13 · ES 10 · TR 7 · SE 6 · CO 5 · AT 5 · UZ 4 · PH 4 · KZ 4 · KR 4 · NO 3 · MX 3 · IL 3 · EE 3 · TH 2 · plus 10 further countries with 1 request each. Sum = 90,681, without a country code 0.
B6 Directory, as of 2026-09-01
| Route | Entry | Status |
|---|---|---|
agent-ready/assess | listed, lastUpdated 2026-08-30T15:27:39.854Z | last settlement of the same URL 2026-08-30T15:27:39.310Z (delta 544 ms) |
bfsg/check, holidays, invoice/check, llmstxt/generate, outreach/policy, deadlines | no longer listed | still listed in July, no rename since, no new settlement |
imprint, readability, iban/check, vat-format/check, robots-lint, x402-demand | not listed | never settled by a foreign payer |
Total 14,681 (pagination.total, cross-checked by exhaustive paging, identical figure).
B7 Proof of the two August payments
0x55a22565f8ec58a6913ca658f1d12f1280f659183bcae1809c39de083f240331— 2026-08-08, 22:32:39 UTC, 0.01 USDC,agent-ready, payer0x1392…3d7aD, country SG, identifiernode.0x322aadec90b772f8c5362f1bfa758fb29254222c835eaacde1e4cbef8ba8a971— 2026-08-30, 15:27:39 UTC, 0.01 USDC,agent-ready, payer0xC533…92D7A, country US, identifiernode.
B8 Metrics, all nine
| Metric | Value |
|---|---|
| Independent buyers | 2 in August (cumulative since measurement began: 4) |
| Conversion, 402 to paid | 2 of 68,280, 0.003% (against real price requests: 2 of 3,832, 0.052%) |
| Repeat-purchase rate | 0% (0 of 4 cumulative buyers has paid a second time) |
| Revenue per resource | 0.02 USDC total, exclusively on agent-ready; all other 12 routes 0 |
| Buyer concentration | 50% per buyer (2 buyers, 1 purchase each) |
| Contribution margin per request | not measured (marginal cost zero, fixed costs not allocated) |
| Error and refund rate | no refunds; 2 of 90,681 foreign requests (0.0022%) ended in a 5xx |
| Availability | not measured in this window |
| Time to first 402 | 2 minutes 50 seconds after the window opened — for a continuously running service this is no longer an activation signal, just the window start, marked [MED] |
