Here's a FEN that can be reached from French defense, Nimzo, Queen's Gambit, and even Sicilian: r1bq1rk1/pp3pp1/2n1pn1p/8/1b1P4/2NB1N1P/PP3PP1/R1BQ1RK1 b - - 0 11
First, thank you for sharing such insightful observations—the visualizations are intuitive and informative.
I had a quick question regarding data access:
From the code, it looks like it makes requests to the endpoint `https://explorer.lichess.ovh/lichess`, specifically to retrieve move statistics (win/draw/loss counts) for given FEN positions.
Does this endpoint query a live Lichess database, or does it return statistics from a precomputed static dataset?
In other words, are the results periodically updated to reflect ongoing games, or are they based on a fixed snapshot? 2) Does the script make a call to the Lichess database ( does not seem like)?
"One might offer a simple explanation: "This makes sense. If a position holds a genuine, objective advantage, more time allows a player to calculate more precisely and maximize it." But I think this is a lazy explanation, and possibly wrong.
To emphasize why this is non-trivial, let's imagine for a second that the curves were inverted, that performance was more extreme in Blitz than in Classical. It would be very easy for me to give a convincing explanation: 'Of course! In Blitz, once a player has an advantage, the defender doesn't have enough time to dismantle it. The chaos of lower time controls naturally exacerbates any edge.'
That explanation sounds perfectly plausible, doesn't it? "
No, not plausible whatsoever. If a position is objectively bad, no amount of calculation can "dismantle" it, unless the players are unbalanced or so weak they can't spot what to do in a position. I don't find this counterargument at all compelling. Let's think about the edge cases: if either player has mate in one, more time on the clock is only going to increase the certainty that the attacker finds it. In faster time controls, chaos indeed blurs the advantage.
I have a strong suspicion that the main trend in all of these plots is sharpness and familiarity. Why some openings lead to larger deviations from the baseline as time goes down should have a lot more to do with how sensitive the position is to suboptimal moves, and this effect is actually less pronounced at lower ratings because those players are missing so many moves *every* position is sharp
Stunning topic. I was looking for a good approach to pratice some data insights skills with my team. So, I will collect a little more data and share with my team at work and come back with some data enginering black magic. Hope to contribute soon. For now, thanks a lot, loved it.
"Since you analyze too broad of openings, you can get so many different positions from e4–c5 that they will have nothing in common in terms of game plan, winning chances, initiative, etc., although still objectively playable. "
* Yes, 1 e4 e6 2 d4 d5 3 exd5, 3 Nc3, 3 Nd2 3 e5 are all French but quite different.
Instead of the French bread logo it were better to put a small diagram of what you mean.
Grasping the 'why' behind a chess move is like finding a master key; it unlocks a door. But knowing what to do after that door is open demands imagination, a trait not equally shared by all. A player's rating measures overall strength, not the source of that strength. Simply playing many openings won't raise your rating; it's about how well you truly understand and play them.
https://lichess.org/training/openings
My current challenge on Lichess is bridging theory and practice. The "Opening Training" section offers puzzles, but it doesn't teach the moves. I have to go to the opening explorer for that. For instance, to learn the "Sicilian Defense: Bowdler Attack," I repeatedly move pieces on the board while saying its name, aiming for short-term memory. Then, I must consistently play it in real games, if my opponent allows, to truly add it to my repertoire. There seems to be no other way.
https://lichess.org/training/Sicilian_Defense_Bowdler_Attack
This struggle shows that mastering complex chess knowledge, like memorizing a phone number, takes real effort and time. Few are gifted with effortless retention; for most, becoming a master is a testament to persistent hard work, not instant memorization.
Dont refer to rating as Elo for crying out loud xDD
I didn't notice anyone else pointing it out,but at least the last graph " have no idea what is going on with the modern defense:"
is just a bad match,
it makes huge waves just to make continuous wave from data.
I'm assuming those dots are data points and lines are showing progress of a function that approximates it.
so that first huge wave is just completely made up just because of variation of data somewhere else.
Exactly threexk, overfit data
Very interesting approach, thanks for the article. I'm wondering how the baseline average elo changes would look like for each time format.