@aseyboldt actually I found the existing example very intuitive and accessible, so I wouldn’t necessarily change it.
But the existence of an apparent bias (due to the confound) and the fact that the sampler has problems, especially when the number of groups increases perfectly set the stage for discussing confounds, reparametrisation, etc. so I’d extend the existing example with sections discusing those thigs or linking to a “second part” that further develops the example and demonstrates how successful pooling with a large number of groups can be facilitated - instead of fitting the paradox into a common example.