GM Ben Finegold notoriously says "Never play f6 [as black, or f3 as white]"

Is this good advice from a statistical perspective? In this project, we analyze 70.315.588 games from the Lichess online database in order to find out:

This website allows you to:

• Pick any move played by white or black, and compare the average score of the games where this move was played against a baseline of the average score for all games.
• Calculate a plot of the average score of your move as a function of the elo range, to see how successfully the move was played by players of various elo
• Filter games and keep only the games where the move was played in the opening, the middle game or the endgame.

Check out this reddit post for a discussion about the results: https://www.reddit.com/r/chess/comments/iv8qxf/update_should_i_play_f6_some_extra_analysis/

The average score S of a move X is: S = (1 * wins + 0.5 * draws + 0 * losses) / (wins + draws + losses)

where:

• wins is the number of games won when playing the move X
• draws is the number of games drawn when playing the move X
• losses is the number of games lost when playing the move X

The part of the game that happens before move 10 is the opening

The part of the game that happens between move 10 and move 40 is the middle game

The part of the game that happens after move 40 is the endgame

⚠️ Note that these three definitions are subjective and highly debatable, for example an engame would be more accurately defined by the point in the game where the queens are traded off, however for the purpose of this project it was simpler to just consider the number of move in order to filter out openings and engames.

Some moves in the Lichess database came with

• Annotations: for example b4! or Kf1??
• Comments: for example e4 [The king pawn opening]
• Computer analysis: for example e4 [+0.23]

All of these were removed to only keep the arithmetic notation.

Additionnally I also chose to deduplicate positional notation (Rfd1 -> Rd1) and captures (Bxf7 -> Bf7 or exd5 -> ed5)

To analyze the first 10.000 games of a PGN stored locally, run:

python main.py lichess_db_standard_rated_2020-07.pgn 10000

To analyze the first 70.000.000 games from a PGN stored in S3, run:

python main.py s3://gjgd-chess/lichess_db_standard_rated_2020-07.pgn 70000000

cd app/
npm install
npm run start

• The script to build the database can be run using Python3
• The webapp was written in React JS and uses the CRA tools
• The deployment to AWS CloudFront is done with the Serverless framework
• The CI/CD is ran using Github Actions.