Guide Statistics 6 min read

White's edge in Lichess's January 2013 file came mostly from the move itself

Moving first gives White a small edge in chess, and we measured it in 121,332 rated games from the Lichess open database's January 2013 file. White scored 52.85%, and about two fifths of that edge came from White being slightly higher-rated on average, which leaves roughly 1.7 points of score for the move.

Nadia Rahim
Data & Statistics Analyst
Published 7 Nov 2026, 10:57 PM (SGT)
Share:
A wooden chessboard with both sides' pieces set up in their starting positions, the light pieces in front. A wooden chessboard with both sides' pieces set up in their starting positions, the light pieces in front. Photo by 17831348 on Pixabay
Advertisement

White moves first in chess, and the value of that head start is usually quoted from master databases. We counted it in the Lichess open database file for January 2013, which holds 121,332 rated online games. White scored 52.85%, counting a draw as half a point. Only 3.28% of the games were drawn. White was also the higher-rated player, by about 10 points on average, and under a plain Elo curve that gap accounts for about two fifths of the edge. That leaves about 1.7 points of score to the move itself. Between players rated fewer than 100 points apart, White scored 52.33% against the 50.14% the ratings alone predict.

For scale, Jeff Sonas's 2002 fit to 266,000 games from 1994 to 2001 gives White an expected 54% at identical ratings. The two agree on direction and little else.

Higher average ratings brighten White's raw score by about a point

We averaged, game by game, the score White should get from the rating gap alone. That expectation came to 51.10%. White averaged 10.2 rating points above Black (median 8) and was the stronger player in 51.63% of games. Of White's 2.85 points above an even score, 1.10 come from the ratings, or 39% on our arithmetic. The remainder, about 1.7 points when only rated games are counted, is what we attribute to moving first.

The split rests on a model. Our expectation is a logistic Elo curve, but the database notes say "The WhiteElo and BlackElo tags contain Glicko2 ratings". The file records no rating deviation, and the Lichess FAQ says new accounts start at 1500 ± 1000. Glicko's own expectation pulls uncertain ratings toward 50%, so ignoring that uncertainty probably overstates the rating share. We found no Lichess page explaining why White was the higher-rated side that month.

White's move still shines through in evenly matched games

The 51,348 games between players fewer than 100 points apart set most of the rating question aside. White won 25,891, drew 1,962 and lost 23,495, a score of 52.33% (95% interval 51.91–52.76) against 50.14% predicted. By our own fit, the move is worth about 15 points on Lichess's Glicko-2 scale (15.7 in these games).

Sonas's figure comes from a straight-line formula fitted to games from Vladimir Perevertkin's database at classical and other long time controls. In score, his model puts White 4 points above even at equal ratings; our near-equal games put White about 2.2 above what the ratings predict. The populations differ too much to read anything into the size of that gap.

Rapid games dim White's score below the blitz figure

We sorted games with Lichess's own FAQ rule, "estimated game duration = (clock initial time in seconds) + 40 × (clock increment)", taking bullet up to 179 seconds, blitz up to 479, rapid up to 1,499 and classical from 1,500.

BandGamesWhite score95% intervalDraws
Bullet32,98652.83%52.30–53.362.63%
Blitz46,28453.57%53.13–54.023.45%
Rapid39,08052.05%51.57–52.543.60%
Classical2,71652.10%50.25–53.953.31%

Blitz and rapid differ: their intervals do not overlap, and among near-equal players the gap holds at 52.93% against 51.61%. Bullet sits between them, so clock length shows no trend. Time forfeits ended 31% of games, and White scored 53.64% in those against 52.49% in normal finishes. We attach no cause to either difference.

Advertisement

Draws stay rare at every rating in the file

Among near-equal games the draw rate roughly doubled with strength, from 2.98% at 1200–1399 to 6.18% at 2000–2199 (712 games), the top band large enough to count. White's score showed no clear trend across the same bands; their intervals overlap.

Over the board, draws are another matter. Steven Brams and Mehmet Ismail, citing Adorján's 2004 book Black is Still OK!, report draw rates rising from 35% for non-elite players to 58% for elite ones at classical time controls.

The 64% quoted for elite play spotlights decisive games alone

Three units circulate as "White's advantage". In our file the score is 52.85%. White's share of the 117,350 decisive games is 52.94%. The raw win rate is 51.21%.

A King's College London release about the Brams and Ismail paper said that "in matches that have a winner at the elite level of human play, white has a winning percentage of 64 per cent". That is a share of decisive games. The paper, arXiv 2108.02547, gives 55% for players rated 2100 and above and 64% for 2700 and above, citing Adorján's book and Chessgames.com statistics on classical games. It also cites a separate 64% from González-Díaz and Palacios-Huerta for expert matches above 2500, and the release does not say which it meant. The nearest comparison to our 52.94% is the 55%, from a different population and era.

FIDE's Laws give repetition draws only to a claim by the player to move

The FIDE Laws of Chess taking effect on 1 January 2023 draw a game on threefold repetition "upon a correct claim by a player having the move", and Article 9.3 treats fifty moves without a capture or pawn move the same way. Article 9.6 ends the game with no claim after a fivefold repetition or 75 such moves, and if the last of those 75 moves gives checkmate, the mate "shall take precedence". The Laws "cover over-the-board play" and postdate our file, whose draws were settled by Lichess's own rules of 2013.

Our international chess board follows those articles in its revision of 7 November 2026. Playing 1. Nf3 Nf6 2. Ng1 Ng8 3. Nf3 Nf6 4. Ng1 Ng8 in pass-and-play mode brings the start position back a third time, and a claim button appears for White. Against the computer it appears only on your own turn. Until that revision the button could also appear while the engine had the move, a loaded PGN game came back with its player tags and result blanked, and a pasted FEN lost its en passant square when copied before any move.

Two limits remain. A claim by intended move, which Articles 9.2.1 and 9.3.1 allow, is not offered. And the only dead positions the board detects are those with insufficient material, so a locked pawn chain plays on. Its move generator, chess.js, matched all 16,996,788 perft nodes we ran from the Chess Programming Wiki's reference positions. How moves are written is covered in our notation guide.

A single month of games illuminates that month alone

The file is lichess_db_standard_rated_2013-01.pgn.zst, listed at 17.8 MB and released under CC0. Its checksum matched the one Lichess publishes. It includes 135 games dated 31 December 2012, which is why we call it the January 2013 file.

Lichess's known-issues list says that before March 2016 "Some players were able to play themselves in rated games". We found no game with the same username on both sides, though a second account would not show.

That month's file holds 121,332 games. Lichess's about page, read on 7 November 2026, says its users play "more than five million games every day", and the database page listed 8,315,708,510 standard rated games. A month from early 2013 speaks for neither that site nor masters at a board.

Advertisement
Nadia Rahim
Data & Statistics Analyst

Nadia Rahim covers statistics, data literacy, measurement, and how published numbers get misread for RECATOOLS.

View author profile → · Editorial policy

About this byline Nadia Rahim is a RECATOOLS editorial persona for statistics and data-literacy coverage. Articles are produced and reviewed under RECATOOLS editorial supervision.

Corrections policy

Advertisement