Comparing chess strength over years and servers
Am I better than my older self?
Introduction
I used to train and play chess seriously, with a view to improvement, in years 2009-2012, and I also do the same now, in 2023-2024. That’s more than ten years difference, so, naturally, I wanted to know how good I was back then and whether I’ve already reached my previous top form by now.
Easy, you say, just look at your rating, which is a relatively reliable measure of a chess player strength. However, I didn’t play many officially rated OTB games back then and thus didn’t have a FIDE rating. Instead, I played online on two chess servers: ChessCube (they closed down in 2020) and FICS (Free Internet Chess Server) which still exists but is much less popular than the current Big Two. And of course, comparing online ratings from two different servers ten years apart is meaningless, because they normally use different rating systems and have different player populations.
Idea of comparing accuracy
So what does one do? I decided to proceed with an idea behind the comparison of top players strength across the whole history (e.g. comparing Capablanca vs Fischer vs Carlsen to answer the ultimate question and determine the GOAT). Namely, the idea of using accuracy of one’s moves as evaluated by the engine. This makes sense, because the engine is considered an ultimate arbiter (Stockfish is currently rated around 3600, if you have doubts). If you play close enough to the engine, then you must be good.
The next question is: how to calculate this accuracy for your own games? I didn’t want to spend much time on this, so I just piggybacked on ChessCom accuracy score: the score it prints after the game when you do Game Review. However, the accuracy varies quite significantly between different game types. I found that in quiet positional games it’s easier to get a high accuracy for my level, but in a sharp tactical game many amateur moves are considered inaccuracies or mistakes, so the score goes down pretty quickly. In order to fight against that I decided to just average the accuracy of many games, thinking that the “law of big numbers” should smooth out irregularities.
When ChessCom introduced their accuracy score system (called “CAPS”) they posted this table:
Note that it’s likely no longer accurate, since “CAPS” was later superseded by “CAPS v2” (and I couldn’t find a similar table for CAPS2). But it still gives a nice illustration of the ranges.
Dealing with time controls
Another problem that complicates the comparison is different time controls. It’s definitely expected that the accuracy of slow games on average will be higher than that of blitz games, so we shouldn’t directly compare them.
In 2011, I was normally playing slower and serious 45+45 (or 30+30) games - I was a member of Monkey Club, a nice friendly chess community, so could afford playing such long games with other members. Later I also played many 15+0 games.
And now I either play 15+10 games on ChessCom or something like 20+5 in rapid OTB tournaments.
So I decided to calculate the accuracy in four categories:
Slow games from 2010 (at least 30 minutes per game)
Rapid games from 2011 (mostly 15+0)
Rapid games from 2023 (mostly 15+10)
OTB games from 2023 (20+5 or 25+5)
Methodology and results
For each of the four categories, I took 30 games and added them to ChessCom library. Then I clicked “Game Review” button on each of those games (which resulted in many browser tabs!), waited until analysis is done, added the resulting numbers to a spreadsheet in Google Docs and averaged them per category.
The results are as follows:
What are the conclusions here?
We can say that the pinnacle of my career so far in terms of accuracy is “Slower online games in 2010”! I almost got to 84.0 accuracy, which corresponds to 1700 OTB Elo according to the CAPS table above. Currently, my just established classical FIDE Elo is 1673, which is fairly close to that. It’s not surprising, given that I like to take my time and 45+45 or 30+30 time controls allow to look deeper into position and make less blunders.
Games without increment in 2011 were rather inaccurate, which can most likely be explained by exactly the lack of increment and more chaotic games as a result, where I am more likely to blunder.
We can also say that my OTB games are more accurate than online games, even though OTB games have less increment (5s OTB vs 10s online). This can be explained either by the fact that I am playing more seriously in rated OTB tournaments, or it may be because I don’t record my last moves in the time scrambles when I get below 2-3 minutes, so the engine doesn’t get to see all the terrible last minute blunders!
And finally, coming back to my original question: have I reached my level of 2010-2012 yet? Unfortunately, it’s hard to say conclusively, because I played different time controls now and then, but in general it looks like at least I am not visibly worse, and most likely of similar strength.
We can also comment on the methodology itself: the results that it produces are quite sensible and the estimated OTB rating for slower games is even close to what I really have now, which boosts my confidence that this exercise wasn’t a waste of time. But of course, there are limitations: just 30 games in each category is probably not enough to make those figures statistically significant, but I believe that they provide a good estimate.
A screenshot of detailed data:
Bonus: useful articles
“Accuracy, Ratings, and GOATs” - an article on ChessCom about using similar methodology to find out who is the greatest player of all time
“Better Than Ratings? Chess.com's New 'CAPS' System” - another article from ChessCom introducing their accuracy score: CAPS
“What is accuracy and how it’s measured” - also tells about the difference between CAPS and CAPS v2 - help page from ChessCom
ChessMetrics - a completely different statistical model to compare historical performance by Jeff Sonas (currently the main FIDE statistician, a person in charge of ratings)
Outro
I hope you’ve found this post useful and/or entertaining. I plan to write to “64 Squares” blog weekly, so please feel free to subscribe on Substack and get new posts delivered directly to your mailbox.