thanks for posting this. It is really helpful and super important!
One thing that currently keeps me busy is the randomization inference with a Difference in Difference estimator.
Idealy I would like to present the single difference estimator, the single difference estimator controlling for previous outcomes (ancova), and the DiD estimator. However, for the latter, of course the treatment group variable changes from 0 to 1 when being in the time during the treatment. This then leads to the error: "b_treat doesnt seem to be constant within clusters".