Alright, here come some more data points: JBB spinning for 24:17.... This is a graph of the raw date minus quite a few values that where high above the curve. I set those to zero by hand.... I am not happy with the data handling program I am using. That is why I did not put much effort in preparing a better graph. I have Libre Calc. Similar to Excel which I also do not like much for this kind of task. I need to get something similar to "Origin" again.
Excel works well for me, but I've been using it for decades. Most of all, I want (1) easy mathematical analysis, (2) easy plotting of the raw SDCs, and (3) easy addition of exponential trendlines. Why the last? Because I find the exponential best fit useful as a measurable reference curve providing an easy way to compare our empirical SDCs. The exponential lifetime (reciprocal of the trendline's decay constant) is a very handy figure of merit for decay rate prior to the SDC tail.
Excel does plotting and exponential trendlines well enough, and with pain I'm willing to live with. Just got Google Sheets to do all 3 things, too, but much more painful and much less flexible.
Not all values are lying on the curve. Mostly this is due to precession and to the traveling of the top, the laser did not always hit it properly. Especially in the later phase of the spin this becomes a problem. So this is another reason why precession in the beginning is not good.
I also had to hand-cull some obvious outliers due to precession and travel. And I can relate to the trouble you had twirling JBB straight into quiet sleep. I get around that with a starter. If I'm just capturing an SDC, why not?
In my experience, the larger the AMI, the smaller scrape angle, the shorter the stem, the larger the tip radius of curvature, or the more concave the support surface, the more tries it takes. It's also important to hold your mouth just right during the twirl.