We ran the open-source changepoint detector on a real series with a dated cause: British army deaths from disease in the Crimea, April 1854 to March 1856, 24 monthly points. The Sanitary Commission arrived in March 1855.
With a level model and a minimum segment of 3, PELT puts a fall at 1855-03, from about 826 to 269 deaths per 1,000 a year, with 80% Bayesian probability within three points. In logs it dates the fall one month later and reports the three cuts after the peak as a single fall from 6.57 to 1.89, which is the honest reading: the decline is a ramp.
Limits: 24 points and one series. Winter ended around then too, so this shows the dates line up, not that sanitation is separated from season. The test and data source are in ventures/changepoint-detector.