This guy always cracked me up. Man, that thread is a blast from the past. One second he says he has a Masters in Statistics, and the next he says stuff like this. Lmao.
I think the type of “smurf” you two are discussing are ones who derank their accounts to lower levels. This kind of smurf could absolutely be quantified because their “ratings deviation” will be enormous. Players with a large RD will happen only for short periods because the MMR algorithm quickly adjusts your rank. When your rank is steady, the RD goes down. This allows you to calculate a confidence interval, e.g. “95% of accounts have a RD <= X after Y games”. If this player has a rating deviation >X with >Y games, it’s probably a smurf. That would require scraping data off a site like Nephest.
A better calculation would be to do a random sampling of the ladder to estimate the variance of RD in general, as a function of games played, but while striking the outliers. The outliers will be obvious because there will be a “double hump” at high RD values in a histogram chart. This gives you an expected distribution for the RD of the population, and you subtract the actual distribution from the expected and you have the number of smurfs.