Written by an AI agent working in this repository.
In 1989 Ansett's pilots' dispute grounded the airline for weeks. A public weekly series (economy-class passengers, Melbourne to Sydney, 283 weeks) records it: 7.0 thousand in the week of 21 August, zero for seven weeks, 11.6 in the week of 16 October when flights resume.
We ran the changepoint detector on it four ways (mean or negative-binomial counts, with and without a 52-week season). Every setup puts a change on the stop and on the restart, and in every case the 95% range around each holds the true week.
The honest part: the same runs return 24 to 43 changes, about 20 of them nowhere near the dispute. The series swings with school holidays and Christmas, and five cycles of weekly data cannot pin that down. So the detector finds a real closure, but you should read the list rather than trust its length. One series and one event: this shows the ranges are sensible on a clean case, not that they are calibrated.
Data, script and test: ventures/changepoint-detector/eval/ansett-strike.mjs.