Long story follows, but Chess.com did not do a good job on puzzle battle's rating algorithm - probably in a rush to get this new feature to players as quick as possible, they just reused the algorithm that calculates ratings in chess games and applied it to battles. It leads to all sorts of problems.
What to do about it? Share the following with everybody that you play with an incorrect rating!
________________________
I have already sent a message to chess.com asking them to correct their program for battle ratings, but they have not fixed it yet. If you also inform them that they aren't using a good algorithm for these puzzle battles, then maybe they will change to something that works better.
This message is circulating among everyone that has an incorrect rating in the hopes that enough of us will contact them.
Please feel free to resend it to anyone else that you have a match with that is not scoring what their rating would have you expect.
Take care!
PS Please read on only if you are interested in understanding why the algorithm gives you a rating that is too low - and how their calculation can be improved.
Here is the explanation:
As it now works, chess.com assigns an ARBITRARY initial rating to members when they start doing puzzle battles - and this leads to many rating problems. Also, as stated in their help pages: "Chess.com uses the Glicko rating system, and part of this system is a number called a ‘rating deviation’ or RD, which measures how sure we are of what your rating is."
This means that they guess a rating for players to begin with when they start doing puzzle battles - and this guess is almost always wrong - and wrong by a lot!
The disadvantages of this system are that:
1) it is very slow in calculating a correct rating because the RD measurement is too slow in correcting their bad initial rating guess, so then
2) it can generate groups of users with too low ratings that then get paired to each other - which then leads to a large number of under-rated players that play against each other and get trapped in a bubble of low rating, and
3) it was designed last century to provide a standard ranking system for players based on match results for chess games, where, as far as calculating performance levels, chess games are in a different category that is very distinct from puzzle battles.
The last point can be understood easily: just completing one puzzle battle gives a very good estimate for a player's strength by using the number of puzzles solved. On the other hand, this is not the case for someone playing just one chess game. Why is this so? Because in a puzzle battle you are really competing against the clock - nothing your opponent does make the puzzles easier or more difficult. Obviously a chess game is different, as what your opponent does can make it more or less difficult to win.
Just a little bit of thought makes it evident that any good algorithm for puzzle battles will quickly give a rating that is proportional to how many puzzles someone can consistently solve. In fact, the overall statistics suggests that multiplying the number of puzzles solved by 50 gives this proportion - now, though, your multiplier is significantly less than this, showing that your rating is too low.
And it's not true that easier and better rating systems do not exist. For example, an alternative and simpler rating system for battles is to use a 3 minute puzzle rush to establish a player's initial rating. How? By multiplying by 50 the number of puzzles solved in those 180 seconds. With this idea you would have started with the rating calculated for you above as your base rating to then get similarly rated members for battles.
As stated before, if you randomly select other puzzle battles for a statistical analysis, you'll notice that this correlation factor of 50 fits the data pretty well. This solves the problem caused by the current system wrongly estimating a player's strength and it also prevents large groups forming of underrated players. (Of course, the statistics should be calculated on players with correct ratings - otherwise lots of garbage data, such as the player with 19 puzzles solved and a rating of only 190, would distort the correlation calculation.)
Another option they have is to continue using the Glicko system but instead of guessing some arbitrary rating - as they currently do - they could apply it to the number of puzzles solved.
I'm not sure if I need to report his, if its a bug or if its by design.
I just played a puzzle battle, and got paired to a 1364 rated player in puzzles.
This while my own rating is only 944.
Offcourse, I got blasted away, and have to say. I didn't find it enjoyfull. It wasn't even close.