Appendix 1—table 3. Accuracy identifying bifurcations in clinical data.

A. Reviewers were highly consistent with each other scoring the bifurcations in the clinical data. (Fleiss Kappa, three reviewers). Each patient was stratified based upon whether their EEG had DC-coupled recordings, as the non-DC group could not disambiguate SN or SH with DC shifts. B. The model fit of the automated features to each chosen bifurcation was performed versus human majority vote, then bootstrapped as in Step 1. The chosen model parameters were clearly highly descriptive of chosen bifurcations in DC onset and offset and non-DC offset. As expected, the lack of DC coupling makes it difficult to disambiguate the four onset bifurcations.

A.Reviewer agreement on human data
	Dc (N = 51)	Non-DC (N = 69)
Onset	p=1.78e-15	p=4.51e-12
Offset	p=4.51e-11	p=9.72e-4
B. Automated features permutation test
	DC	Non-DC
Onset	p<1e-4	p=0.0969
Offset	p=7e-4	p<1e-4