That's a good question. I should create a new video for dealing with that. I haven't looked into this issue, but presumably you would want to treat the return to baseline as a totally separate phase for analytic purposes because you'd expect that there is some level of residual that persists after the removal of treatment so it isn't the same as baseline. For this reason, you should not treat the return to baseline as simply "more baseline". David Barlow has excellent texts on the topic, but they were published before some of these statistics were developed. Mind you, I'm not an expert in single case design methodologies but simply wanted to clarify some basics for testing simple designs. Historically these have been assessed in a non-statistical manner (i.e., visual assessment of graphs), and the various statistical approaches are fairly new (within the last 10-20 years). By way of comparison, Gossett devised the t-test in 1908, and by that time Pearson had been using the correlation for about 10 years (having adopted/stolen it from his mentor Francis Galton in the 1890s)