All systems answering

Components

Each part behind an answer, reported separately. A premium check that was never switched on for this deployment reads "not configured" rather than degraded, and is not counted against the headline above.

Availability, last 90 days

100.00% of the 22,206 requests recorded in the last 90 days were answered without a fault on our side. 28 of those 90 days recorded at least one request; 62 recorded none. 68 of those requests were answered, 22,138 were refused as answers about the request, and 0 failed here.

Incidents

Written by hand, so this page carries what we found and wrote down: the strip above counts requests and cannot say why a day was bad. Each report says what went wrong, who it reached and what changed afterwards. The Atom feed at /status.rss carries the same entries.

• Four datasets stopped refreshing and nobody was told

  Four datasets behind the answers had stopped refreshing, the oldest copy dating from 6 September, and the daily check that exists to catch exactly that reached nobody. A fifth was flagged by a freshness window set too tight.

  16 Oct 2026

• Two datasets had stopped syncing and nothing said so

  Two weekly data syncs were failing, and because a scheduled background task throws its output away, the failures left no trace anywhere. The datasets were missing from the status page before anyone noticed.

  8 Sep 2026

• IP answers lost their location and network flags

  Files a scheduled job had synced correctly were written into directories the web process could not enter, so IP lookups answered with no location and no network operator, and the datacenter, VPN and Tor flags read false.

  5 Sep 2026

• Documentation and content pages answered 500 in production

  A class the app loads at runtime was only installed for development, so every documentation page, content page and the sitemap failed on the production server while the lookup endpoints kept answering.

  3 Sep 2026

The full incident history

Measured latency

Time spent inside the service per request, recorded after each response is sent. It excludes the network between you and us. Percentiles cover the requests we answered: a refused key, a rate limit or a validation error returns in a millisecond or two and is no evidence that the API is fast, so refusals are counted in the success rate instead. Groups appear once they have at least 20 requests in the window.

Measured quality

Everything above says whether we answered, and how fast. These say something harder and much smaller: how often the answer was right. Each product is measured a different way, because the honest measurement available differs by product, and each one below states what it is counted from, over what window, and what it cannot tell you. None of them is an accuracy rate, and we do not have one. They are measured once a week rather than live, so each carries the date it was taken, and they are not part of the JSON above.

Data freshness

Every answer names the datasets it used; this is the version each one carries. Where the publisher stamps its own release date in the file — a census snapshot, a quarterly register — that date is shown, so an old date can still be the newest release there is; otherwise it is the day we last built or synced the copy. Feed groups show the oldest sync date among the feeds they download, so the whole group is at least that fresh; a list that ships with the app is not synced, so it dates a group only when nothing else in that group is. A register published in several files, such as Brazil's 27 states, is counted the same way and shows a date only once every file has synced. Each row is judged against the same freshness window the daily sweep alerts us on, set from the sync's own schedule plus, where the date is the publisher's, the publisher's release cycle: for a copy dated by our own sync, current means the latest scheduled run landed, and a single missed run turns the row stale here rather than leaving a date you have to check yourself. A list we curate by hand — the sanctions and FATF lists, the satellite and mobile networks, the address formats — is judged against a review window instead, and reads stale once its check is overdue; the FATF lists' window counts from the plenary they reflect rather than the day we checked them, so they turn stale soon after the next plenary however recently they were checked. A version that is not a date — a library's semver, a publisher's release number — carries no age and is marked "not dated" rather than stale.