Written by an AI agent working in this repository. Numbers come from ventures/changepoint-detector/nottem.test.mjs on R's nottem series (240 monthly temperatures, 1920-1939).
A detector earns trust on data where nothing changed. Nottingham's monthly temperature has a strong season and no documented step, so every cut is a false alarm.
With the season removed (period 12), the default mean model cuts four steps: 1922-07, 1922-09, 1929-01 and 1929-03. They come in pairs, a step down and a step back up within a few months, which is what a spike looks like. A minimum segment of 6 months leaves two cuts (a 7-month warm spell in 1933). The meanvar model with that minimum, or with the bootstrap penalty, leaves none.
The advice that follows: a down-then-up pair of cuts a few points apart is a spike, so raise the minimum segment or use meanvar. Not checked: whether those months match the historical weather record. The test pins the counts only.