Coverage and accuracy

What Tokyo Events holds, where each event comes from, and how often we check it. Every event links to its original listing and records when we first saw it, last saw it and last fetched its source. Spec: openapi.json. Service status: status. Changes: changelog.

10,897 upcoming events from 101 sources; data published Nov 6, 2026. 124 of them are listed by more than one source and shown once. Machine-readable copy: data/coverage.json.

Known gaps

Where this site is thinner than it should be, so you know what not to rely on it for.

• Some kinds of events are much thinner than others. Music, classes, family events, talks and theatre are well covered; food and drink, games, volunteering and nonprofit events, networking, conferences, comedy, film, science and outdoors have far fewer listings than a city this size should.
• The largest Japanese ticket sites are not in yet: e+ and Lawson Ticket do not answer automated requests, and Ticket Pia listings carry a date and venue but no start time. Concerts and shows from those sellers are under-counted.
• Peatix, the biggest community-event platform, is only partly loaded so far, and connpass (tech meetups) is not connected, so community and tech meetups are under-counted.
• Several city and library calendars block automated reading, and a few ask not to be read by AI crawlers at all; we honor that, so events that only appear there are missing.
• Listing text stays in the language of its source (mostly Japanese); there is no translation.

How fresh

Each source is fetched on its own schedule. Of 101 sources with upcoming events, 9 were fetched within 2 days, 99 within 7 days, and 2 are older than 7 days. Oldest last fetch: Oct 23, 2026; newest: Nov 6, 2026.

How many real events we catch

What these numbers do and don’t say: recall covers only sources that publish a complete listing we can re-read; for the rest we have no completeness figure. The comparison uses our own filtering rules, so it shows the fetch lost nothing, not that every kind of event is included. Accuracy is a random sample; pages that block automated checks or need JavaScript are not counted. The audit is a snapshot and is repeated, not continuous.

By source