How the research works
The calendar fills itself for the most part. This is how – and how it stays honest.
Every night
- Reading. The crawler TangoKalenderBot reads the public event pages of tango schools, venues and listing calendars – politely, at most one page per second, by the rules of
robots.txt. Pages that do not change are read less often. - Extracting. An AI model pulls date, time, venue, title and repetition rule out of every page – and only what the page says. Every find comes with the passage of the page as evidence; without evidence nothing is taken over.
- Matching. Every find is matched against known venues, organizers, series and events. An event that two pages name independently is the same event with two sources – not two events.
- Publishing. The sources make a level: Hint (one source), Corroborated (two independent sources or the organizer's own website), Corroborated externally (a partner calendar that checks with the organizers), Confirmed (the organizer). Hints nobody confirms fade out by themselves.
- Telling. Organizers whose events we found get exactly one mail with a link to take them over – never a second one without a reply.
What the sources are worth
Every source has a weight that follows from its hit rate every week: when organizers confirm a source's finds, its weight rises; when they contradict them, it sinks to zero. New sources the research finds itself are on probation for a month – their finds count as hints only.
A golden set of known pages checks every week whether the extraction still works precisely. When precision drops, no event climbs a level until it is right again.
What we do not do
- No texts, no pictures from other sites – the calendar shows data and a short summary of its own, naming and linking the source.
- No signing in, no forms, no page over two megabytes.
- The text of a page read is kept only until it is extracted, a week at most.
- Whoever does not want to be listed says so once – it holds permanently.
The event page of every found event shows the sources and, field by field, where the details come from. The technical details are on the page TangoKalenderBot.