I Analyzed Over 1000 Games & Here's What I Learned 

I'm happy to see you making use of the Training Program database. It would be nice if you pick some pieces of analysis from time to time (maybe not full games, just positions) and comment on the submitted analysis : good, bad, incomplete, how to improve, etc. It would certainly be useful for the students.

This is an extremely hard data set to analyze and you were right to be cautious. The biggest issue is that it's observational data. In other words the openings weren't randomly assigned. Given that each player selected their own openings, and as white they dictated the choice while with black they were reactive, I wonder if there was an effect of player choice that's hidden. For example, perhaps bookish players choose the English, which prioritizes positional understanding, plus there's a familiarity effect. Also, if there are multiple games from the same players over a longer period of time, that's a time series. Meaning you would expect someone playing an opening the 10th time to do better than the 1st time, and probably be higher rated if they are under 1600. You've done a pilot study, which has generated some interesting hypotheses. Nice job.

As a young player around the expert range in USCF I can confirm me and a lot of my friends hate playing against 1.d4

Interesting data and very impressed of such a high number of games submitted in the relative short time for program has been up and going.

Hey Kostya! Thanks for the analysis! Quick comment: I think that determining White Expected Score with White Average Rating vs Black average rating can be slightly misleading, as it should be the average of the expected scores for each game (say your pool consists of three White players, rated 2800, 1500 and 1500; and three Black players, all rated 1600. White average rating is 1933, Black average rating is 1600, so we would expect White to score 1.74 out of 3 points (around 58%), but looking only at the average ELO we would predict White to score 2.64 out of 3 points (around 88%)). Not sure how relevant the effect is for this pool of players, though. Thanks again for the insights!

This is great info. We appreciate your hard work.

I love theese guys so much they have a really special content keep going guys❤❤

This was an outstanding video. I am now subscribed solely on this content. This highlights that the game changes with level as evolution of players happens. Thanks

A note on errors; standard deviations (assuming win/draw/loss have equal chances) are about 1.2% for n=1000, about 4% for n=100 and about 12% for n=10. So a hundred game sanple (say the Vienna) could just have a 3% (2SD) chance of being +8% or more. Given you present over 20 results, probably one or two are a bit wild.

Scientific review I'm used to on RU-vid. Bahaha!!! 🤣🤣🤣

from my personal experiance playing many openings over my climb to ~1950 online, I think most of this data is reliable and should closely reflect what you should expect trying out different openings and studying them. I used to play e4 exclusively, and what terrified of the idea of touching the cursed d pawn on move 1. I was kinda "1.e4 best by test" kinda player, and would play pretty much anything slightly unusual and gambitlike in 1.e4 e5: Danish, Kings gambit, Vienna, Scotch gambit, Evans, Knight Attack and so on. At the time, I thought 1...e5 was just a joke that plays straight into my hands, And I just used it as a sandbox, choosing different hyper-aggressive offbeat lines to crush opponents early on. Later, I started experimenting with 1.d4, building a Catalan based repertoire where the bishop usually goes to g2 and you play a long game. In the beginning, I didn't really know how to treat the positions and would try to force through unsound exciting attacks. Overtime, I started realizing the real merits of playing 1.d4, Which for me is just to keep pieces on the board and keeping a long term structural advantage. I found that oftentimes with the kings pawn, and specifically 1.e4 e5, pieces were falling off the board, leading to early dry positions where White is equal or even flat out worse. But, with the Queens pawn, I could very easily keep my precious pieces and grind my opponent down, playing a long game that oftentimes even gave me my beloved attack at the very right moment. As for playing Black, I am surprised the French scores so well in low level, as I tried playing it myself with no more than decent success (I guess either the French isn't for me, or more likely I was studying it wrong). At the same time, I really hated facing the French with White so... And I truly believe the Caro in every level. I think most Caro players at low levels just don't quite play the right lines that give you this real dynamic Caro (specifically 3...c5 in the Advance). I would reccomend the Caro to antone struggeling against 1.e4 under my level. I would also like to adress the Sicilian statistics. In the video, it was said that the Sicilian basically slways scores somewhat poorly for Black at all levels, but I think that is a little misleading. Concidering the best try for White is the Open Sicilian, I would assume some Sicilians are just born better than others. What Sicilians are played? Dragon? Najdorf? Sveshnikov? Kalashnikov? Classical? Taimanov? This is the magic of the Sicilian, it is usually black who chooses the variations, unlike 1.e4 e5. And as for the Anti Sicilians, I am a very big believer of the Morra, And I think people usually just play it for tricks and traps, lowering the scores for those who know their stuff. The main problem is that it doesn't have much independent value... which sucks... I full heartly reccoment Marc Essermans book on the Morra, "Mayhem in the Morra". Amazing book. Lastly, I think first move sidelines like 1.f4 can be played with overwhelming succses using the right research, and again it is shadowed by those who just play it wrong. This is my general view. I think statistics are a really powerful tool that should be used togrther with engine analysis to build a killer repertoire for both colors. Usually, a sound idea that scores poorly can be made practical by just digging a little deeper. and of course, statistics on non commital moves like 1.e4, 1.Nf3 and 1...c5 shouldn't be taken seriously as the important decisions were not yet taken. Hope this was helpful for somebody reading this. Go to sleep.

Wow you put in a lot of work on this. Thanks Kostya!!

It would be great to know the biggest mistakes in each game (in terms of centipawns) and analyze that data...

I haven't even submitted my games to skew the open Sicilian stats with my 86.4% winrate in it. Open Sicilian by far the best Sicilian

Would be interesting to do a follow-up now, with 6 more months of data

Oh, hey! The Pirc shows up (though no Benoni). Nice.

outstanding, .. thank you, .. m.

I would expect that the Alekhine scores better than all other defences against 1.e4, as it is rarely played and can be tricky for white.

And here I am too lazy to analyze more than 5 games lol

fascinating

Right off the bat I would like to point how bad the science is on this. ;-) And what I learned is that I, as the 1000 rated player, should DEFINITELY switch to the English opening. :-P

"... the typical scientific rigor you see on youtube." :-)

Can't wait to see the same analysis when you get to 5000 or 10000 games.

Interesting. Though I'm glad my games aren't poisoning the sampling! The summary @18:55 was worded the wrong way round. Or maybe that's a test to see if we're paying attention. 😉

This video is kinda useless since you didn't adjust the ratings on each platform. The Lichess -> FIDE difference is so massive that you may as well not even do this. Saying that you "aren't trained" in mathematics doesn't excuse sloppy work.

Interesting 🤔

Is Kostya 'a professional chess player with a rating of about 2400 at the age of 29' ? XD

Absolutely meaningless. Anyone who comes away from this video concluding that the English is better than e4 for 1500 players is completely misguided.