Ratings Are Broken

@A_Kireev said in #89:
> Therefore, it would be logical for both you with K = 20 and the master with K=10 to count the games with owners of K= 40 with a reduced K (for example, 5). A stable rating obtained in the fight against the holders of the same stable rating should not decrease sharply due to the fact that the opponent's rating is provisional.

Here I fully agree. My proposal was a bit different by looking at TPR instead of rating for people having no stable rating yet. People with a provisional rating/ rating based on few games harm too much the other ratings with the current system.

what is the difference between many Ks, and factors known to correlate (do we need causation, in things already constructed out of correlations). There is no causation other than the game outcomes in the ratings system.. everything else is hidden.

what is the difference between obscure K factor denomination and having age structure in the model.

also someone said one rating system for the whole population. would you like to hear my leitmotiv? One population distribution assumption with very low dimensionality to start with being K factor chiseled one at a time?

please correct my mis-conceptions about that understanding i have about ELO as in fide using such limited freedom of shape population distribution. otherwise, it screams.

It looks like the browser extension Lichess Custom Stats that I mentioned in prior posts is not using bullet games.
At the moment, it seems like we need to play at least rapid games to have a stat visible by that extension.
If not this is what's visible.

Quote: NaN% (0)
Avg Opponent: 0