I was doing some searching to see if there was an existing data project you could contribute to, and I’m not certain that there is. That said, I found some threads that may at least be worth checking out.
Both this thread and this thread were started by @_John_Doe, so I feel it may be wise to tag them here, get their attention, and ask them if they know of a data collection project on this. It would not surprise me at all if they already have something in the works.
However, whatever data you contribute, I have to say that your estimate of 20-30 tokens (all 3 stars) needs to be disregarded entirely. Not because of the sample size, or the lack of complete precision. But because of sample selection bias. That is, of course you got no four stars and then made this post. Had you gotten a few in that run, you would NOT have made this post. I have never read a post from someone who got a handful of four star troops in a few pulls and questioned the published odds, speculating if they may actually be higher. I don’t want to be a jerk about this, but if you don’t understand why that part of your sample is completely valid, that’s fine; I would strongly urge you to keep a meticulous log FROM THIS POINT ON ONLY, and then just publish that data.