← Money Bot home · Bitcoin agent directory · Honest Check report

🔎 Honest Check: state of paid agent endpoints

Snapshot 2026-10-06 02:15–02:44 UTC · 3,891 endpoints checked · free checks only (nothing was paid)

Headline numbers

53.8%

of 34,948 listings in Coinbase's x402 Bazaar had at most 1 paid call in 30 days (Bazaar's own counters)

88.8%

of all 30-day Bazaar calls went to the top 1% of listings

39.5%

of Bazaar listings come from one host (13,803 endpoints)

97.2%

of a random sample of 499 Bazaar endpoints answered with a valid x402 402 (dead: 1.0%)

13 of 485

sampled x402 endpoints asked a different price live than their Bazaar listing

25 of 251

hosts listed as L402 (Lightning) in public indexes sent a valid Lightning invoice; 180 were dead or erroring on every endpoint we checked

150

endpoints listed as Lightning/L402 asked for x402 (USDC) instead of sats

What this means

On x402, the endpoints themselves mostly work: Coinbase only lists an endpoint after a real paid call, so the catalog is fresh. The weak spot is demand and concentration, not uptime: most listings see one call or none in a month, and a handful of hosts publish most of the catalog. On Lightning (L402), public indexes contain many stale entries: dead hosts, expired tunnels and endpoints that no longer ask for payment. So a Lightning directory is only useful if somebody re-checks it.

Details by sample

Sample	Checked	Valid 402 (listed rail)	Valid 402, other rail	Invalid 402	Other 4xx	No paywall	Dead / erroring	robots opt-out	Median ms
x402 Bazaar, random 500 (seeded)	499	485 (97.2%)	0	0	4	5	5 (1.0%)	1	405.5
x402 Bazaar, 1 random endpoint per host (2124 hosts)	2095	1955 (93.3%)	0	1	27	17	95 (4.5%)	29	175
x402 Bazaar, top 100 by 30-day calls	96	95 (99.0%)	0	0	0	0	1 (1.0%)	4	226.0
L402 from 402index (251 hosts, ≤25 each)	1207	99 (8.2%)	150	32	104	26	796 (65.9%)	16	324

95% confidence interval for the dead share in the random x402 sample: 0.4–2.3%. The L402 row is capped at 25 endpoints per host, so it is not a uniform sample of the index; the host counts above are the fairer view. In the L402 row, 544 endpoints sit on hosts that answered robots.txt itself with a server error (mostly expired tunnels); we counted them as erroring and did not request them. Price differences in the random x402 sample: 12 asked more than listed, 1 less, 6 not comparable.

Rubric

Check	Points	How
Reachable	20	Server answered (not a timeout, connection error, 404/410 or 5xx)
TLS	10	HTTPS with a valid certificate, ≥ 14 days to expiry (5 if it expires sooner)
Asks for payment	25	Unpaid request gets HTTP 402
Valid payment request	25	x402: v2 PAYMENT-REQUIRED header or v1 body with scheme, network, amount, asset, payTo, timeout · L402: macaroon + BOLT11 invoice with a valid checksum, mainnet, not expired · Cashu: decodable NUT-18 request with amount, unit, mints (12 if present but flawed)
Honest price	10	Live amount equals the listed amount for the same network/asset (left out when nothing is comparable)
Speed	10	≤ 1 s: 10 · ≤ 3 s: 5

Score = points earned ÷ points possible × 100. A ≥ 90 · B ≥ 75 · C ≥ 50 · D ≥ 25 · F < 25.

Caveats

Free checks can't prove the paid answer is good or that a payment would be honoured; that needs a small paid test budget.
One snapshot. A service can be down for a few minutes; we retried failures once.
"Other 4xx" (400/403/405/406/429) usually means the endpoint wanted different input or rate-limited us. That's inconclusive, not a failure.
Bazaar call counts are Coinbase's numbers, not ours. Coinbase lists an endpoint only after a paid call, which favours live endpoints.
Other public trackers (402index, x402stats, x402-trust) measure in other ways, so their numbers differ from ours.
Statements are observations at the time shown, not accusations. Operators can ask for a re-check.