Week 106: when to castle queenside, learned from three bad examples

Eight games between February 16 and 22, 2026: 2 wins, 6 losses, rating 1260 → 1228 (−32). With White I went 1–5. When I looked for the common thread, the engine handed it to me three times: in three different games, all with White, it marked the same move as a blunder — O-O-O, castling queenside. So this week's question is simple: when to castle queenside, and when the king on e1 is doing a job that nobody else will do once it leaves.

The week in numbers

None of the eight games ended in checkmate; six ended by abandonment, and all six were my losses. My two wins came with very different accuracy: 95.0 % against Zeus_22_0 (1187) and 47.2 % against Misha1995285 (1222), a game full of blunders on both sides that ended when my opponent resigned on move 15.

When to castle queenside: game 1980

Against Gent71 (1267) I had my usual London-type setup, with the queen on b3 and the h-file half open after Black took my bishop on g3.

r . b q . r k .
. p p . . p p .
p . n . p n . p
. . . p . . . .
. . . P . . . .
. Q P . P N P .
P P . N . P P .
R . . . K B . R

Castling looks like the safe, natural move. The engine disagrees (+0.35 → −2.39), and its reason is in the first line after 10.O-O-O: 10...Ng4, a knight move that attacks my f2 pawn, which now has no defender, and after 11.Re1 Nxf2 the knight attacks my rook in the corner (−2.21). With 10.Be2 the game stays about equal (+0.37). My opponent played 10...Bd7 instead, but the next move I grabbed a pawn with 11.Qxb7, another blunder, and the game ended by abandonment on move 33 (game).

The same idea, twice more

It would be one bad game if it hadn't repeated. On Sunday, February 22, I played two London-style games with Black's knight sitting on e4, and in both I castled queenside:

• Against CheesyChessChannel (1260), 12.O-O-O was answered by 12...Nxf2, a knight that takes the pawn and attacks my rook on d1 and my rook on h1 at once — a fork (+0.60 → −3.15). The engine preferred 12.g4 or 12.Bd3, both slightly better for me. The game ended right there.
• Against chessoccer10 (1248), same picture: 10.O-O-O, 10...Nxf2, the knight attacks both rooks (+1.01 → −3.57). 10.a4 kept me slightly better. That game also ended on move 10.

The pattern is the f2 pawn. Before castling, my king on e1 defends it. After O-O-O the king is on c1, my rook arrives on d1 and the other rook still sits on h1, and a knight on f2 hits both. A knight on e4 is one jump from f2, so with a black knight there, castling queenside is a gift.

A different kind of castling lesson: game 1979

The cruellest example came from the other side. Against Im-not-AI-and-i-know-it (1278) my rook had gone to b7 and Black's bishop on c3 was attacking my knight on d2, which was pinned to my king.

r . . q k . . r
p R . . p p . p
. . n . . . p .
. . p . . . . .
. . . . . . . .
. . b . P . P .
P . P N B P P .
. . . Q K . . R

14.Bb5 pins Black's knight on c6 and looks aggressive. But Black simply takes, 14...Bxd2+, with check (−2.96). If I recapture with the queen, the engine's line shows the trick: 15.Qxd2 Qxd2+ 16.Kxd2 and then 16...O-O-O+ — Black castles queenside with check on my king, and next move the king takes my rook on b7 (−7.08). 14.Rb3, bringing the rook back to attack the bishop, kept me slightly better (+0.54). I lost that game on move 61 (game).

The lesson of the week

Castling queenside is not automatically "king safety". It moves the king two squares away from f2 and puts my rook on d1, on the same row as the rook on h1. So before O-O-O I now check three things: is there an enemy knight that can reach f2 (from e4, g4, d3 or h3)? Is anything else only protected by my king? And can the move be met by a check that wins material? In games 1980, 1983 and 1985 the answer to the first question was yes, and the engine's quiet alternative — Be2, g4, a4 — was always the better move.

All games this week

Previous: week 105 · Next: week 107 · Journey hub · Review your own games like this