One important principle in all this is that we're looking for *differences* between a sample heartbeat and some norm. So far, we've been doing comparisons by the normal a/b method: listen to one, then listen to another; look at one waveform, then look at another. However, we could create a picture which incorporated not only the sample data, but the differences between it and a typical dataset. For example, if a typical waveform were this:
