I found the links in the last few posts very interesting. Thanks.

In the Forumslader instructions, I see five electrolytic capacitors. I would expect one to be connected for the reservoir/smoothing function after the rectifier,
leaving four on the AC side to cancel inductance. Assuming the common polarised parts, they need to be connected in series pairs. Those traces can be seen in the
board layout diagram. So my interpretation is that there are only 2 or 3 speed-matching stages. No visible relay, so it would be interesting to know how the switching is done.

Steve (steelbikeguy) wrote:
"I've generated this sort of model for my 1st generation Schmidt by loading it with various resistances and riding at different speeds. With a large matrix of that data, it is possible to figure out the values that make the model match the data.
On the plus side, the measured value of dynamo resistance matches the model pretty much exactly.
I never tried to measure the inductance, so I can't comment on how well that might match the value derived from the testing over a range of speeds and loads."

I did something similar, as well as measuring resistance and inductance directly when stopped. I have a poor selection of resistors, so not a large matrix.
The results match the higher of the the static inductance measurements, but the resistance came out 50% higher than the static value (6 ohms).
That might be due to magnetic/eddy losses, but I would not claim much accuracy for the measurements.

"I will say that riding up and down the street, over and over, with a meter strapped to your handlebars, is bound to get you some strange looks. I recommend a quiet back street."
I would not like to try that with the local potholes! I turned the bike upside-down and drove the tyre from the chuck of a variable-speed power drill, then a small grinding wheel
for higher speeds.

"I think I've seen the core losses modelled like that before, but I don't think I understand how the test data reveals the value."
I think I recall trying to match figures from a website (perhaps Steve's) using that two-resistor model. The results were worse (higher residuals) than the simple one,
with a clear systematic bias.