You can also automate the measurement of the break ration, although I did not find a suitable filter in Audacity, although the function for the first step of that is available.

The principle is the same that most analog break/make measuring instruments used by Western Electric, for example in the No. 2B test sets for pulse train measurements.

The method involves clipping your sound recordings at the top and the bottom to a square wave with the zero line exactly in the middle. That means removing any DC bias from the signal. Let's say you clip the data to values +1 and –1, centered on 0. This can be done with the clipping function in Audacity. In addition the data should be clipped to a some number of complete dial pulses, not more and not less. I am showing the result below for your No. 2 sample.

The next step is to divide all data into sample slots, which they are in already actually, because the data is sample at a certain frequency. I believe it was 44.1 kHz in your samples. So, all you have to do now, is compute the sum of all samples, i.e. adding the sequence of –1 and +1 values. Image that the make/break ratio is exactly 50%, this would mean that there are the same number of negative values as there are positive values and the sum would be zero. For 66% there are 2x as many samples of one polarity vs the other. There the sum is a direct representation of the break/make ratio. With the polarity in your data, negative being a break, the sum is positive below 50% and negative above 50%. All you need to do is count the sample properly which is the 100% value and normalize the value.

This should be very simple to implement in an Audacity plug-in, the hardest part being learning how to do that. I would actually be surprised this does exist already somewhere as a third party module. I will look.

~~I will attach the promised image in a moment...~~

The top graph shows the original data trimmed to 9 pulses, and the lower half shows the clipped data normalized to +1 and –1.

BTW, if one uses this clipped signal, the correlation function cleans up amazingly, and provides a nice bell curve for reading the principle frequency.