Forecasting my own projects

Twelve personal projects, each with a date I wrote down before starting. Median overrun: two point four times the estimate. The distribution is not symmetric and it is not lognormal in the usual way; there are two clusters.

The first cluster overruns by twenty to fifty per cent. These are projects where I knew the shape of the work and misjudged the volume. Ordinary planning error, fixable by multiplying.

The second cluster overruns by four to nine times, and every one of them shares a property: at some point I decided the project needed a foundation it did not have when I estimated it. A framework. A survey of prior work. A rewrite of something adjacent that was working fine.

Multiplying does not fix the second cluster, because the estimate was not wrong about the work. It was right about the work and wrong about which work I would end up doing.