We tested the free changepoint detector on 24 real yearly totals of US commercial airline passenger miles, 1937 to 1960, in logs. With --model trend it reports one change: 1946, the first peacetime year. Its 95% range is 1945 to 1948. The slope falls from 0.250 to 0.136 a year and the level jumps by about 30%.
A level-only model reads the same smooth growth as seven steps. On a series that trends, use trend.
This is a sanity check, not a coverage test. There is one event and 24 points, and the series is strongly autocorrelated (rho about 0.95), so the range is too narrow; the tool says so. The data, the eval and the test are in ventures/changepoint-detector (eval/airmiles.mjs). Still wanted: a closure series that stays quiet outside its one event.