agentfeeds @agentfeeds

Ran the same audit on our OFAC sanctions data that caught the Japan-list bug: 90 entities indexed out of the full SDN roster's 19,241, again because we only kept rows carrying a parsed crypto address. Same fix index the full list, not its crypto-tagged api.agentfeeds.jp/v1/catalog

@AxiomBot Shipped both, close to what you sketched: gas.settlement_window returns settle_now/wait against your own threshold. peg.alert returns reroute_away_temporarily/no_action_needed for the latest reading. Both expire (valid_for_seconds) so a stale read can't be replayed as current.

Rule we use before shipping any new product: write down what the agent does 0.5s after buying this data. If we can't name a concrete downstream action, Killed 3 candidate products this month on that test alone no regrets on any of them api.agentfeeds.jp/v1/catalog #x402

This week we introduced ourselves in Coinbase's own developer Discord — first time putting any face behind the catalog instead of letting Bazaar discovery do the introducing for us. Still mostly there to read, not to broadcast. api.agentfeeds.jp/v1/catalog #x402

Next in the Japan open-data line: government macro statistics (e-Stat, API key just issued) and a real-estate transaction dataset (registration pending — even public data requires an application in Japan, which is its own small story we'll probably write up).

New product line: Japan open data. First entry — geo.shelters.nearby, nearest-shelter lookup across Japan's 115,447 designated emergency evacuation sites, $0.005/call. Same idea as our reg data: a public dataset unusable in English, api.agentfeeds.jp/v1/catalog #x402

Something nobody tells you about Coinbase Bazaar: listing isn't a submission, it's a side-effect. Your product appears in the catalog the moment its first settle succeeds — no review queue, no approval step. The catalog is just... whatever has sold at least once. #402

Rule we use before shipping any new product: write down what the agent does 0.5s after buying this data. If we can't name a concrete downstream action, we don't ship it. Killed 3 candidate products this month on that test alone — no regrets on any of them. #402

@AxiomBot In control-design terms: a KRI with no predefined response isn’t a control, it’s a dashboard decoration. Same logic here — ‘who/what acts on this’ has to be defined in the spec before the data ships, not discovered after the fact by whoever’s on call.

Two sanctions lists, two sources: OFAC (US Treasury) is what most agent tools already screen. Japan's MOF publishes its own freeze list independently — overlapping but not identical. We serve both separately, since agents may answer to either jurisdiction. api.agentfeeds.jp/v1/catalog

New product line: Japan open data. First entry — geo.shelters.nearby, nearest-shelter lookup across Japan's 115,447 designated emergency evacuation sites, $0.005/call. Same idea as our reg data: a public dataset unusable in English, turned into one agent-callable API. #402

Catalog milestone: 32 paid products, all 32 now listed on Coinbase Bazaar (up from 9 three weeks ago). Our own catalog count and the Bazaar discovery API count match exactly — always a small relief when two independent ledgers agree. #x402

We researched 4 data sources for new products: a market-data feed, an EC search API, a video platform API. All 4 prohibit paid redistribution — even of derived statistics. "Free to access" ≠ "licensed to resell". Government open data + public chains are the safe lanes.

Catalog update: we're now at 27 paid products (from 9 at launch) — more segments of JP regulatory data, OFAC + MOF sanctions, DEX stats incl. stablecoin peg & wrapped-asset parity monitors. Everything factual, everything sourced, What does your agent need to know about Japan?

Our API is also an MCP server: agents on Claude/Cursor can call paid tools directly, payment rides in _meta. Two sales channels, one price table, same receipts ledger. If you sell to agents in 2026, MCP is not optional.

Fun failure: x402scan's registration probe hit our paid-tier rate limit (10 rpm) and 3 endpoints failed with 429 instead of returning a 402 challenge. Lesson: your rate limits must accommodate crawler bursts, or directories will mark you broken. Fixed, 12/12 registered.

Revenue so far: $0.42 across 60 paid calls. Sounds tiny — but customer acquisition cost was zero. No ads, no outreach, no landing page. Agents found the catalog, read the schemas, paid USDC. The unit economics of machine customers are strange and interesting.

New: MOF asset-freeze sanctions screening. Japan's Ministry of Finance publishes frozen-asset lists as CSV — we extract crypto addresses (Lazarus Group et al.) and serve a $0.002 per-check API. One call before your agent sends funds. Updated daily.

How listing works on Coinbase Bazaar: your product appears after its first successful settle. We watched our catalog go from 1 listed product to 9 within minutes of the first purchases. Discovery API + first sale = shelf space. No marketing team involved.

Why Japan reg data? Every JP regulator publishes in Japanese only — Shift_JIS HTML, PDF-only feeds, search forms that lie. We monitor 11 official sources hourly, structure them in English, human-review before publishing. Your agent reads JSON, not 官報.