The x402 Machine-Payments Marketplace, October 2026: The Coinbase Discovery Catalog Three Months On

Snapshot date: 2026-10-21. Source: the Coinbase CDP x402 discovery API (https://api.cdp.coinbase.com/platform/v2/x402/discovery/resources), retrieved in full — 28,428 listed resources in 29 paginated requests (limit=1000). Previous snapshot: 2026-07-08, 24,927 resources (the July report in this section). Section 8 adds the discovery catalogs of the eight other facilitators that expose one — 94,702 listings in total, 50,909 unique resources.

This report follows the structure of the July study and answers the same questions — what AI agents pay for over x402, how the marketplace is structured, what separates listings with traffic from the rest, how ranking works, where the catalog is saturated and where it is empty — with one addition: every section states what changed in three months. The short version: the catalog turned over almost completely, paid volume tripled, demand moved from Twitter data to inference and transaction execution, and the spam domain of July was replaced by a self-paying long tail.

The catalogs behind this report can be browsed live on x402.cp0x.com: the facilitator list with a search across all nine discovery catalogs at once, and each facilitator's catalog as agents see it — descriptions, tags, input schemas, prices, 30-day activity.

Key takeaways since July

1. Methodology

• The full catalog was downloaded via paginated requests (limit=1000&offset=N); completeness was verified against pagination.total; no duplicate resource URLs were found. The pipeline (download, normalisation, digests) lives in scripts/facilitators/ of this site's repository and was re-run on the July dump, so the two snapshots are compared with identical code. July figures quoted here therefore differ slightly from the July report, which was written by hand on the raw dump.
• Popularity is measured by the catalog's own quality block: l30DaysUniquePayers (unique paying wallets over the trailing 30 days) as the primary metric and l30DaysTotalCalls as the secondary one.
• Three traffic-shape heuristics are applied to every listing:
  • wallet farming — payers ≥ 50 and calls / payers < 1.2 (hundreds of one-call "payers");
  • keep-alive self-payment — payers = 1 and calls ≤ 2, the cheapest way to satisfy a delisting rule;
  • organic reuse — payers ≥ 3 and calls / payers ≥ 3, several independent clients each coming back.
• Revenue is estimated as listed price × calls and is an upper bound: the catalog shows the maximum price (amount / upto), not settled amounts.
• Categorisation: keyword classification over the active segment, then manual review of the top-100 listings by payers and by calls.
• The other facilitators (§8) were downloaded through their own /discovery/resources endpoints on the same day (GoPlausible and P2Flux on 2026-10-22) and normalised to one record per listing; overlap is measured by exact resource URL and by hostname. On-chain settlement figures come from x402-list.com/api/v1/facilitators (30-day USDC settlements attributed to known settler addresses, CC BY 4.0) on 2026-10-21.

2. Market snapshot

Finding 1: the catalog is now a 30-day activity index, not a directory. In July 43% of listings had no paid call in a month; today 60 listings out of 28,428 do. One listing in the catalog spells out the rule it sells insurance against: "resources are delisted after ~30 days without a settled payment" (and offers a $190/year keep-alive). Every listing you see has been paid for at least once since late September. The calls > 0 filter that made sense in July is meaningless now — 99.8% pass it.

Finding 2: volume tripled, concentration doubled. Paid calls went from 233k to 825k per month, but the top resource alone now carries 29.5% of them — an AI "token verdict" endpoint (www.ax1.vc, $0.02) with 243,580 calls from 2,708 payers — and the top 10 hosts carry 73% of all calls (68% in July). The organic-reuse segment — 1,272 listings with ≥ 3 payers and ≥ 3 calls per payer — accounts for 83% of all calls (56% in July). The growth is real; it is also narrow.

Networks. Base is still on 99% of listings (28,148), but the single-chain era is over: 48% of listings now accept payment on more than one network (25% in July). Solana appears on 11,210 listings (6,204 in July), Polygon on 7,417 (1,510), Arbitrum on 5,878 (1,373), and a long tail of new chains has arrived — XRPL (1,856), Monad (1,851), World Chain (1,328), Optimism (1,260), the Circle Gateway chain eip155:5042 (1,012), BNB (933), Algorand (892), X Layer (884), HyperEVM (841). Pure Base-only listings fell to 14,447 (51%); Solana-only listings remain a curiosity (106). x402 v2 is now 99.3% of the catalog (43% in July); 152 listings are typed mcp rather than http (none in July).

Pricing. The median listed price is $0.01. The largest bucket moved up: $0.01–0.10 per call now holds 15,123 listings, $0.001–0.01 holds 10,827, $0.10–1 holds 2,146, $1–10 holds 287 and 35 listings charge $10 or more. Within the active segment (≥ 5 payers or ≥ 20 calls) the median is $0.0085 and the 90th percentile $0.05.

3. Where the demand is: categories

Active segment (≥ 5 payers or ≥ 20 calls in 30 days): 2,428 resources on 826 hosts (1,215 resources on 233 hosts in July).

1. The killer app was replaced. July's leader — a Twitter/X data provider with 52k calls from 177 payers, 48% of all calls in the category of search and scraping — is down to 5,054 calls from 107 payers. Search and scraping fell from the largest category by calls to third place with 7% of the total.
2. Inference is now a volume category. blockrun.ai alone serves 103k paid calls a month (chat completions at $0.002, Exa-backed answers, video generation), and several resellers of Venice/OpenAI-compatible endpoints sit behind it. In July inference was ~15k calls; today it is 106k and the category with the highest estimated revenue.
3. Agents pay to move money and to talk to each other. The two highest-call non-data endpoints are execution fees for signed transfers (keyring-agent, 29k calls at $0.05) and a Chainlink workflow runner (26k calls from 2 payers). An agent-only messaging network (agent.pocket.network/v1/meadow) took 14k calls from 35 payers — the kind of repeat usage that only real integrations produce.
4. Real-world goods appeared. Bitrefill's invoice endpoint (gift cards, mobile top-ups, 180+ countries) recorded 62 paid calls from 20 payers at a listed maximum of $1,000; Laso sells prepaid cards and ACH payouts at $5–100. These listings dominate the nominal revenue table while being a rounding error in calls.

Volume and money still live in different places. Crypto data generates 356k calls for ~$7.4k; the four text-processing endpoints of www.api-xpay.com priced at $10–20 produce $7.3k nominal from 659 calls (see §6 on why that number should be read with suspicion); Bitrefill's $62k nominal comes from 62 calls. In July the revenue leader among pure APIs was a premium-API reseller at ~$715/month; today it is an inference reseller.

4. Provider archetypes

The six patterns from July are still visible, with changed weights, and three new ones have emerged.

1. The killer app, replaced. July's Twitter/X data provider (52k calls, 177 payers) is at 5,054 calls from 107 payers. The new #1 is a single consumer-facing endpoint (www.ax1.vc "token verdict", $0.02) with 2,708 payers and 243,580 calls: 90 calls per payer, entertainment rather than infrastructure, but by far the largest flow of USDC through the catalog.
2. The premium-API reseller, consolidating. stableenrich.dev lost payers (706 → 422) but doubled calls (16k → 29k): fewer, heavier clients. blockrun.ai went the other way — payers 664 → 1,266, calls 10.7k → 103k — by adding inference to the Exa/Firecrawl proxy menu.
3. The data hub. api.onesource.io is still the #1 host by unique payers (7,825 across 70 endpoints, up from 25 endpoints) and still shows the farming signature (§6); it added an order-book exchange and a reserve protocol on Robinhood Chain as new endpoint families.
4. The micro-endpoint swarm. Unchanged in kind, but the top-30 by payers is no longer 25/30 RPC wrappers; a consumer app, utilities and FX ticks broke in.
5. The long-tail carpet, now self-watering. The biggest carpets changed names (lowpaymentfee.com's 10,028 unpaid listings are gone; saylorinnovations.com 1,568, agentshelf.syntexa.ch 1,236, dns.intel.rallylive.ca 694 arrived) and changed behaviour: each listing receives exactly one paid call a month, from one wallet — see §6.
6. The brand. Nansen, Exa, Apify, Chainlink, Pocket Network, Bitrefill: still getting traffic with short descriptions.
7. New: the inference reseller. OpenAI-compatible chat/completions behind x402 — blockrun.ai, api.gondola-ai.com, cheaptokens.ai (sells Venice API credits for $1) — the fastest-growing archetype by calls; price competition is already at $0.002 per call.
8. New: the agent-infrastructure vendor. Services whose customers are other x402 services: an API recommender that is paid discovery (api.agents-trust.com/recommend, 8,353 calls), a 402-response linter (quartermaster.surewhynot.app), pre-payment risk checks (cyre.dev, agent-paycheck), prompt-injection detection, inference notarisation, an "agent-readiness package" at $450, and a $190/year keep-alive. 192 listings now describe pre-flight / safety checks (76 in July); 311 mention MCP (66 in July).
9. New: the storefront. Gift-card and e-gift endpoints at fixed denominations ($5–$100), ACH and card issuance, art sales — few calls, real prices.

5. What makes a listing discoverable: description semantics

July's playbook — task → response fields → parameters with an example → data source → use cases → keyword tail → "no API key, pay per call" → sibling endpoints — was adopted wholesale. Median description length went from 56 to 209 characters; 31% of listings (8,796) exceed 300 characters; 68% carry tags (23% in July); 66% set serviceName; 100% ship a Bazaar extensions block with an input schema (the facilitator now requires it). Keyword tails are pervasive ("defi tvl, protocol tvl, total value locked…").

The consequence is that copy no longer separates winners. Listings with ≥ 21 payers have a shorter median description (115 characters) than the catalog as a whole (209), and only 42% of them carry tags. The leaders are leaders because of brand, a consumer surface, or an integration that calls them thousands of times — not because of embedding-optimised prose. Good copy has become table stakes for the long tail; it is no longer a ranking edge. Two practices still correlate with repeat usage in the top-100: (a) the description states the exact price and billing unit ("$0.002/URL — total = urls.length × $0.002", "pay per second"), and (b) it names the deterministic contract ("Deterministic per ruleset", "Replayable with one GET", signed receipts).

6. Ranking mechanics and metric gaming

• Listing is payment-driven and now expiry-driven. A resource appears when real payments flow through the facilitator with discovery metadata attached, and disappears after ~30 days without a settled payment. 21,536 of July's 24,927 listings are gone, 10,737 of which still had paid calls in July — they simply did not get paid again for a month. Only 13.6% of July's catalog survived.
• Ranking approximates 30-day unique payers, which remains gameable at the cost of gas. The wallet-farming signature (≥ 50 payers, < 1.2 calls per payer) matches 53 resources with 7,195 "payers" and 7,364 calls; 46 of the 53 belong to api.onesource.io, whose endpoints fell from ~370 to ~230 payers each since July but still show one call per wallet. Two more clusters look alike: www.api-xpay.com (9 endpoints at ~215 payers each, 7k calls each, plus $10–$20 text endpoints where payers ≈ calls) and tick.hugen.tokyo. In July the farmed cluster held 24,185 "payers" (58% of the catalog's payer metric by a domain-level count); the stricter signature today matches 7,195 — under 10%.
• The keep-alive economy is the bigger distortion. 11,426 listings (40.2% of the catalog) have exactly one payer and at most two calls (27.6% in July); 12,609 (44.4%) have exactly one payer. On the largest carpets the ratio is mechanical: agentshelf.syntexa.ch 1,235 of 1,236 listings, dns.intel.rallylive.ca 694 of 694, email.intel.rallylive.ca 275 of 275, agents.driftflight.com 417 of 441, market.datapackvibe.com 472 of 545. The 30-day rule removed the unpaid carpet and replaced it with a self-paid one; a provider with 1,000 listings keeps them all listed for roughly 1,000 × $0.01 + gas per month.

Practical reading of the quality block today: treat payers = 1 as "unverified", payers ≥ 50 ∧ calls ≈ payers as "farmed", and payers ≥ 3 ∧ calls / payers ≥ 3 as the only clean signal. That organic segment is 1,272 resources — 4.5% of the catalog — and carries 83% of paid calls.

7. What changed since 2026-07-08

Where July's leaders went. The machine-mailbox search endpoint that led by payers (872) is gone. The 24 api.onesource.io RPC endpoints that filled the rest of the top-25 are all still listed with 40–60% fewer payers each (e.g. tx/:hash 399 → 185, nft-owner 361 → 97). api.tcgapi.dev collapsed from 1,219 listings and 1,224 payers to 5 and 13; 2s.io from 1,431 payers to 284; proxy.suverse.io from 889 to 43; api.interzoid.com from 431 to 53. agent402.tools cut its listings from 1,344 to 613 and lost 300 payers but gained 7k calls. stabletravel.dev lost payers and gained calls (958 → 6,196).

Who arrived. By payers: saylorinnovations.com (2,825, carpet), www.ax1.vc (2,708, one endpoint), www.api-xpay.com (2,463), api.vextorium.com (1,671), agentshelf.syntexa.ch (1,236, carpet), cloud.trycorpus.ai (1,075 payers on 29 cloud-pricing endpoints — the cleanest new B2B entrant), x402.orthogonal.com (616 payers, 6k calls), agent.pocket.network (582 payers, 48k calls). By calls: www.ax1.vc, blockrun.ai, www.api-xpay.com, agent.pocket.network, x402.sniperx.fun (34.6k), keyring-agent.blockchhub.link (29.2k), api.agents-trust.com (8.4k), api.losbeto.xyz (7.3k), api.solenrich.com (69 → 6,815), api.gondola-ai.com (4.5k).

Shape. The catalog went from one dominant dead carpet (one host held 40% of July's listings) to 49 hosts with 100+ listings each, none above 6%. It is both more diverse and more mechanical. Hosts doubled (1,323 → 2,711) and so did resources with meaningful demand (≥ 21 payers, 121 → 228): the ratio of sellers to buyers that the July report called an inverted marketplace is where it was.

8. Beyond Coinbase: the other facilitator catalogs

Nine facilitators answer GET {base}/discovery/resources without a key. Downloaded in full, they hold 94,702 listings that collapse to 50,909 unique resource URLs on 4,129 hosts; Coinbase holds 56% of the unique resources.

ThirdWeb is a mirror, not a second catalog. Its metadata.source field says where each listing came from: coinbase 28,076 (98.8% of the Coinbase catalog), payai 7,651, dexter 1,945, goplausible 1,734. Only the Coinbase-sourced items carry a quality block, and it is Coinbase's. The format is x402 v1 throughout, descriptions are truncated to about 200 characters, there are no tags, serviceName or schemas, and the page size is capped at 100 (395 requests for a full download). With GoPlausible now polled directly, ThirdWeb lists zero resources that no other catalog has, and its own settlement rail did 79 transactions in 30 days. It remains a convenient one-call union in a single v1 shape; it is not evidence of demand.

PayAI is an append-only history, not a state. 22,014 listings, no activity metrics at all, nothing ever removed (the oldest entries date to October 2025). 43% of it is testnet — 9,500 listings accept payment only on Base Sepolia. 48% of it is one provider: the *.intel.rallylive.ca family holds 10,637 micro-utility listings at $0.01. 575 listings carry a price of zero. What PayAI does well is copy and tooling: 87% described (median 248 characters), 92 MCP tools with input and output schemas, and the only non-Coinbase catalog with a working /discovery/search. Raw PayAI counts overstate the Solana/Base ecosystem several-fold.

Dexter is the curated, scored crawl. 3,177 listings on 536 hosts, re-crawled daily (2,058 updated within 24 hours of the snapshot), all x402 v2, and the only catalog with an editorial layer instead of usage: qualityScore 0–100 (median 88), verificationStatus — 2,595 pass, 547 fail, 31 inconclusive — a tier and a nine-value category taxonomy. 117 of the failing listings are simultaneously listed on Coinbase: the only independent way we found to spot dead endpoints that still receive keep-alive payments. Dexter also has the widest network arrays in the set and is the only catalog with Stellar (252 listings). Too small to measure demand; the right tool for liveness.

GoPlausible is the Algorand catalog — and it was wrongly reported as unreachable. The facilitator at facilitator.goplausible.xyz answers /discovery/resources without a key (1,000 items per page, three requests). 2,681 listings (2,656 unique URLs, 25 duplicates) on 234 hosts from 223 merchants; 2,393 accept USDC on Algorand mainnet, 332 on Algorand testnet, 76 on Base, 23 on Solana — 2,589 listings are Algorand-only. The format is Bazaar-like (resourceUrl, accepts, a discoveryInfo block with an input example and an output example on every listing) but without tags, serviceName or x402 version; median description 173 characters. What makes it worth polling is its own activity counter: every listing carries a lifetime settleCount, plus firstSeen and lastSeen. The counters sum to 967,805 settlements; 2,598 listings have at least one, the median is 3, and 245 have 100 or more. The distribution is extreme even by Coinbase standards: one endpoint — onestepchess.xyz/api/v1/moves, a chess-move API at $0.01 — holds 493,189 settlements, 51% of the catalog's lifetime total; x402.twit.sh/tweets/search has 129,091 (it is listed on five of the nine catalogs); six api.syraa.fun market-data endpoints have ~180k between them. The registry grew in a burst — firstSeen by month: June 25, July 675, August 1,063, September 712, October 206 — and is not pruned at 30 days: 1,221 listings were last seen 30–90 days ago and 40 more than 90 days ago. Largest hosts by listings are carpets familiar from Coinbase (agent402.tools 631, sikho-ai.onrender.com 287, vead.app 197, algorandtracker.com 174). Overlap with Coinbase is 611 resources on 10 hosts (555 of them agent402.tools); 233 resources exist nowhere else. ThirdWeb's goplausible feed carries 1,734 of the 2,656 — the mirror is a partial one. x402-list.com does not track GoPlausible's settler addresses, so there is no external settlement figure to compare its counters against.

Ultravioleta DAO is the self-registered registry with telemetry. All 1,530 listings are source: self_registered, and the facilitator then probes them: HTTP status (1,526 answered 402), latency (median 2,040 ms), uptime, an ownership attestation (13 attested, 1,512 not), priceFreshness (88 listings where the listed price disagrees with the probe), and firstSeen (880 listings appeared in the last week). 935 of its resources exist nowhere else; only 15 hosts overlap with Coinbase. The biggest hosts are agent-tooling vendors (x402broker.com 269, agentdiet.dev 268) — the archetype of §4 item 8, listed here before it has traffic. Its usage counters show 83 calls across the catalog in 30 days while x402-list.com attributes 23,565 settlements to its addresses in the same window; usage is not a demand signal yet.

P2Flux, OpenX402, OpenFacilitator. P2Flux (api.p2flux.com/x402, no key, page size capped at 100) lists 58 resources on 6 hosts: five WordPress sites selling individual posts at $0.02 and one sports-statistics API, all Base, all updated within a day, no activity metrics — paid web content rather than APIs, a new genre but a tiny one. OpenX402 lists 47 resources from 3 hosts without descriptions. OpenFacilitator still answers in its legacy shape, has not seen a call since 2026-05-15, and 74 of its 100 listings are Solana.

Overlap. Exact-URL overlap with Coinbase: ThirdWeb 28,076, PayAI 4,190, Dexter 1,203, GoPlausible 611, Ultravioleta 175, OpenFacilitator 9, OpenX402 and P2Flux 0. Hosts present on five of the nine catalogs — x402.twit.sh, agent402.tools, api.agentstools.dev, api.xona-agent.com, x402factory.ai, api.x402node.dev, api.syraa.fun, the x402-bazaar-rank worker — are tooling vendors and carpets that register everywhere; once a 402 response carries discovery metadata, being listed on several facilitators costs nothing, so multi-listing is a property of the protocol, not a sign of traction.

Catalog size versus settlement. Listing counts and money move independently. On 2026-10-21 x402-list.com attributed these 30-day settlements to facilitator addresses: Polygon Facilitator 3,776,469 tx / $111,862 (no catalog); Coinbase 2,194,860 / $616,704; FluxA 222,968 / $744,425 (no Bazaar catalog — see below); Figment 183,242 / $607 (none); PayAI 148,086 / $10,848; Celo 70,289 / $127,127 (none); Dexter 30,525 / $1,934; Ultravioleta 23,565 / $5,475; Meridian 3,472 / $115,657 (none); OpenFacilitator 2,558 / $18; OpenX402 1,486 / $1,273; ThirdWeb 79 / $153. The facilitators with a catalog account for about 36% of measured settlements and about 36% of settled dollars. Polygon and Celo expose /supported but no discovery; Meridian exposes neither. FluxA, the largest facilitator by dollars, runs a facilitator at facilitator.fluxapay.xyz with /supported, /verify and /settle and no /discovery, but publishes its own marketplace catalog at monetize.fluxapay.xyz/api/discover — 472 verified APIs in a non-Bazaar shape (slug, name, tags, price range, access through its router), no activity metrics, no key required. Questflow needs a Bearer token for discovery; KAMIYO and the x402scan "Auto" proxy have shut down. A discovery-based view of the x402 economy therefore sees Coinbase well, the smaller catalogs dimly, and roughly two thirds of the payments not at all: every market-size figure in this report is a floor.

What is usable for statistics. Demand (30-day calls and payers): Coinbase only, with the farming and keep-alive caveats of §6; GoPlausible's lifetime settleCount is the one other usage counter, not time-windowed. Liveness: Dexter (daily verificationStatus), Ultravioleta (health probes), Coinbase (the 30-day rule), GoPlausible (lastSeen). Quality opinion: Dexter and Ultravioleta. Provider copy for semantic analysis: Coinbase, PayAI, Ultravioleta. Network coverage: Stellar only on Dexter, Algorand on GoPlausible, Solana depth and MCP tools on PayAI. Search: Coinbase and PayAI. A practical polling plan: Coinbase monthly for demand and churn; Dexter and Ultravioleta weekly for liveness and new providers; GoPlausible monthly for Algorand and its settle counters; PayAI monthly with testnet filtered out; ThirdWeb no longer needed; P2Flux, OpenX402 and OpenFacilitator quarterly.

9. Saturated versus underserved

Saturated (supply far ahead of demand):

• Single-purpose micro-utilities (unit conversion, holiday calendars, DNS/TLS lookups, string transforms) — now ~10k listings on Coinbase and another 10k on PayAI, almost all with one self-paid call.
• Basic token prices, gas, balances, ENS — a dozen providers per query; the farming cluster sits here.
• Gift-card storefront clones at fixed denominations — dozens of near-identical listings with 0–3 calls.
• "Token verdict / roast / risk score" for memecoins — the #1 app spawned imitators with single-digit payers.

Real demand with proven leaders:

• Inference and OpenAI-compatible proxies (106k calls) — price competition is already at $0.002 per call.
• Web search and page reading for RAG (61k calls) — Exa, Firecrawl, managed unlocking.
• Smart-money / whale / KOL flow on tokens (x402.sniperx.fun, 35k calls from 50 payers).
• Agent messaging and lookup (14k calls on one endpoint).
• Enrichment ($0.15–0.28 per call; 7k calls) and cloud-price intelligence (1,075 payers).

Underserved (visible demand, thin or immature supply):

• Payment safety for agents. Pre-flight checks on counterparties, 402-response linting, spend-shape grading: 192 listings describe it, the best ones have 5–26 payers. The demand side is every agent that pays strangers; the supply side is hobby-sized.
• Paid discovery itself. api.agents-trust.com/recommend (8,353 calls) and a ranking seller (x402-bazaar-rank) show that agents will pay to be told what to call; the facilitator's own search is free but unranked beyond payer counts that §6 shows are gamed.
• Deep DeFi position analytics (health factors, liquidation candidates, vault governance) — still only listed as single endpoints with single-digit payers, while position-level risk keywords appear in hundreds of descriptions. Unchanged since July.
• Compliance primitives — sanctions screening endpoints exist on four or five hosts; none exceeds ~40 payers; the KYC/AML vocabulary is in 230+ listings.
• Mid-priced composed answers ($0.02–0.10). The active segment's price distribution still has a gap between $0.001–0.01 commodity data and $0.15+ enrichment; cloud.trycorpus.ai ($0.05, 1,075 payers) is the proof that the gap can be filled.
• Execution with accountability. Signed-transfer fees and workflow runners already clear 55k calls a month from a handful of payers; attestable, auditable execution (receipts, notarisation) is listed by two or three providers with under ten payers each.

10. Conclusions

1. The Coinbase catalog is now a rolling 30-day index of paid resources: 28,428 listings, 60 of them without a paid call, 13.6% survivors from July. Snapshot comparisons must account for forced churn.
2. Paid volume tripled to 825k calls a month and got more concentrated: one endpoint carries 29.5%, the top 100 carry 78%, the organic-reuse segment of 1,272 listings carries 83%. Estimated revenue grew less, to ~$27k, because the new volume is cheap.
3. The delisting rule created a keep-alive economy — 40% of listings show exactly one payer and ≤ 2 calls — on top of the wallet farming that still marks the #1 host by payers but shrank from 58% of the payer metric to under 10%. payers ≥ 3 ∧ calls/payers ≥ 3 is the only trustworthy public signal.
4. Demand broadened beyond data: inference resale, agent messaging, transaction execution and real-world purchases (gift cards, cards, ACH) are new volume or revenue categories; the Twitter-data leader of July lost 90% of its calls.
5. Listing copy converged on the July template (median description 56 → 209 characters, tags 23% → 68%) and stopped being a differentiator; leaders are shorter-described and win on brand, surface or integration.
6. Multi-chain is the norm (48% of listings accept several networks, nine new chains), x402 v2 is universal, and mcp-typed listings have appeared.
7. Across nine discoverable facilitators the de-duplicated x402 universe is 50,909 resources on 4,129 hosts. Coinbase is the only catalog with 30-day usage; GoPlausible — reachable directly, contrary to our earlier finding — adds lifetime settle counts and the Algorand side; ThirdWeb is a mirror with nothing unique left; PayAI is 43% testnet and 48% one provider; Dexter and Ultravioleta are the only independent liveness checks.
8. Discovery covers about a third of measured x402 settlements. Polygon, FluxA, Celo and Meridian have no Bazaar catalog (FluxA has its own, non-Bazaar one), so any "size of the x402 market" estimate from Bazaar data is a lower bound.
9. The clearest whitespace is agent payment safety, ranked discovery, position-level DeFi risk, compliance, and mid-priced composed answers — the same gaps as July, now with measurable early demand.