The opening move an old book prints as P-K4 is e4 in algebraic, because descriptive notation names each square from the side of the player making the move. In our count of 808 games by Paul Morphy and José Raúl Capablanca, played between 1848 and 1941, 716 contained at least one move that algebraic notation can only write with an extra character, such as the b in Nbd2. That is nearly nine games in ten, though it covers two players from one transcription source and is no modern rate. FIDE recognises "only one system of notation, the Algebraic System" for its own tournaments and matches, and a scoresheet kept in any other system "may not be used as evidence in cases where normally the scoresheet of a player is used for that purpose".
Descriptive notation insists on each player's own side of the board
Descriptive notation names each file after the piece that starts on it. From the queen's side across, they run QR, QN, QB, Q, K, KB, KN and KR, and ranks are counted 1 to 8 from the mover's own back row. So one square has two names. The square White calls KB3 is f3; Black's KB3 is f6. P-K4 is e4 when White plays it and e5 when Black does, and PxP says only that a pawn took a pawn.
Algebraic notation fixes one name per square for both players: files a to h from White's left, ranks 1 to 8 from White's side. Wikipedia's article on descriptive notation already sets out this contrast and FIDE's evidence rule.
Arbiters warn any player writing moves in another system
This rule is from the FIDE Laws of Chess approved by the General Assembly on 7 August 2022 at the 93rd Congress in Chennai and applied from 1 January 2023, current when we checked on 14 October 2026. Article 8.1.1.1 has players record moves "in the algebraic notation (Appendix C)". The preamble to Appendix C adds that an arbiter who sees another system in use "should warn the player of this requirement". For chess literature and periodicals, FIDE only recommends algebraic.
Several symbols are optional. Articles C.13.3 to C.13.6 cover x for a capture, + for check, ++ or # for mate, and e.p. for an en passant capture. That last one changed: the Laws in force before 1 July 2014 said "'e.p.' is appended to the notation", and the edition applied from that date made it "may be appended".
Appendix C prints one sample game three ways. In the 2023 text, the long-form rendering plays 8.Qd4d3 where the other two play 8.Qe3+. Both are legal, so from move 8 the long form describes a different game, without the check. The 2017 and 2018 editions printed that move as Qd4e3+, consistent with the rest, so the slip arrived with the current text.
Game-file notation concedes nothing to FIDE on castling and promotion
Game files in the PGN format follow a standard, revised on 12 March 1994 and coordinated by Steven J. Edwards, that defines its own Standard Algebraic Notation (SAN). FIDE writes castling with the digit zero, 0-0; PGN's export format uses the capital letter O, warning that a zero "can also confuse parsing algorithms". FIDE writes a promotion as d8Q, and PGN as d8=Q. PGN's export format requires x and the check and mate signs that FIDE makes optional, and it gives en passant no special marking.
The two algebraics differ in other ways too. FIDE accepts a long form naming the starting square, such as Ng1f3, marks a draw offer as (=), and allows local piece letters; PGN's export format admits none of them. These symbols are required for the export format; the import format, the standard says, is "somewhat more relaxed".
Master games from Morphy and Capablanca demand an extra character once in 25 plies
We replayed the 808 games from PGN Mentor's two player files and asked, for every ply (one move by one side), whether another piece of the same type could legally reach the same square. Pinned pieces do not count, as PGN's rule says. Of 60,412 plies, 2,405 needed the extra character, 3.98%.
Nearly all of those were by file, as in Nbd2: 2,273 plies. The other 132 needed a rank digit, as in R1e2. None needed the full starting square, a third step PGN provides and Appendix C lacks. A rerun in a second chess library, python-chess, matched every figure. We found no earlier published per-move count; forum users converting old books report that software stumbles on ambiguous moves.
Until 14 October 2026 our chess notation converter wrote descriptive without that extra information, so in our sweep 2,706 of the 60,412 plies came out as a string another legal move would also produce. It wrote a bishop taking a knight as BxN6, could not read descriptive at all, and failed at the first header line when a whole PGN game was pasted, in all 808 games. The revision of that date reads descriptive as well as algebraic, including Kt, Castles and ch, and writes a qualified form such as N(QB3)-QN5 only when a plainer one would also fit another legal move.
En passant and promotion each claim roughly a game in 24
En passant came to 36 plies across our 60,412, in 34 of the 808 games. Promotion matched it almost exactly: 36 plies in 34 games, 11 of them made by capturing. That is about one game in 24 of this corpus for each, by our arithmetic. A FIDE scoresheet may write exd6 e.p. and d8Q; a PGN file writes exd6 and d8=Q.
FIDE recommends figurines for printed periodicals
Appendix C.3 lets each player use the piece names "commonly used in his/her country", giving F for the French fou and L for the Dutch loper, and recommends figurines, small piece symbols, in printed periodicals. PGN calls figurine notation FAN and says the two notations are "otherwise identical".
PGN lists alternative piece letters for 17 languages, English included; all are European and none is Asian. The standard reserves these letters for local display software, not for use in archived files. Chinese chess writes its moves in systems of its own, compared in our xiangqi notation guide.
Capablanca's 1921 book admits only Kt for the knight
Capablanca's Chess Fundamentals, copyright 1921, on Project Gutenberg, shows descriptive as it was printed. In it, Kt marks both the knight and the knight's file, so K - Kt 1 is a king move. The book never uses N for a knight. Checks appear as ch, 295 times in our scan, and never as +.
Re-editions carry a risk of their own. Edward Winter, writing about a 1994 algebraic edition of Capablanca's My Chess Career, reports that "mistakes have been added", among them a game result printed the wrong way round.
Chess Fundamentals also often drops the K or Q in front of a flank file. Our regex scan, which is approximate, found 1,584 of 2,878 non-capture moves naming a bishop, knight or rook file bare, as in P - B 5: 55%. Only 382 qualified the file in full, as in P - K Kt 4. In the bare form, the board position is the only way to know which flank is meant.
View author profile → · Editorial policy
