Castling short looks like the most dangerous move in chess, and the truth is stranger: it is a trick of how we write moves down. O-O is the one move with a symbol of its own, while every knight move is split across a dozen spellings, so castling wins a count it was never really in. Group blunders by the PIECE that moved instead and castling is the smallest category there is.

Castling short looks like the most dangerous move in chess, and the truth is stranger: it is a trick of how we write moves down.

O-O is the one move with a symbol of its own, while every knight move is split across a dozen spellings, so castling wins a count it was never really in. Group blunders by the PIECE that moved instead and castling is the smallest category there is.

O-O tops the list for 60 of 239 players and is about 1.1% of any one player's blunders, measured across 305,509 games containing a move that cost at least 300 centipawns. Read it as a fact about counting, not as advice: newsroom_corpus_worst_piece() puts castling at 1.4% of blunders against the queen's 19.8%.

Across 305,509 analysed games from 239 players.

The Club Ledger · No. 2

Getting winning is not the same as staying winning.

For the typical player in the corpus, 25.0% of the games in which they reached an advantage of at least three pawns finished with the evaluation back at level or worse, measured over 242 players and 290,319 such games.

The middle half of players sit between 20.4% and 30.9%, so this is a habit nearly everyone has rather than a few people dragging an average. Pooling every game instead gives 25.5%, near enough the median that no one prolific account is carrying it. Only positions from move five onward count as the peak, so opening noise cannot qualify as winning, and the endpoint is the final evaluation rather than the result, which also turns on the clock.

Across 290,319 analysed games from 242 players.

The Club Ledger runs again tomorrow.