The chess fortress: how to save endgames when you are material down

Short answer

A fortress is the position where extra material produces nothing: no king entry, no pawn break, no zugzwang that makes the barricade give way. This guide covers the three patterns club players actually meet — the rook-pawn corner with the wrong-coloured bishop, rook and pawn against queen, and opposite-coloured bishops — checks each of them against the endgame tablebase, and marks the exact line where each one stops working. It also covers the case where the engine's evaluation bar disagrees with the exact result.

The essentials in a few lines

• Black king on h8, white pawn on h6 and a light-squared bishop is an exact tablebase draw — and Stockfish 18 shows +0.84 in that same position at depth 30. This is the case where the evaluation bar lies.
• The corner fortress hangs on a single tempo: with the black king on f8 and Black to move, 1...Kg8! draws; with White to move, 1.h7! wins. Same position, two results.
• One spare tempo destroys the barricade: add a white pawn on g5 to that position and it becomes mate in five — 1.Bf5 Kg8 2.Kf6 Kh8 3.g6 Kg8 4.h7+ Kh8 5.g7#.
• Rook and pawn on f2 against a queen is a tablebase draw: the queen has no move that forces progress, and Stockfish agrees, showing 0.00.
• With opposite-coloured bishops, two extra pawns draw while the passers stay within two files of each other (b5 and e5) and win at three (b5 and f5) — the king blocks one pawn, the bishop watches the other's queening square, and at three files apart one of them is left unguarded.

What it is, and what it is not

The definition matters because it is operational rather than decorative. A difficult defensive position that you hold for thirty moves and then lose is not a fortress: it is a defence. A fortress has a verifiable property — the defender owns a set of waiting moves that can be repeated forever, and no sequence by the opponent breaks the blockade. When that property is present the result is half a point, and the material count is irrelevant.

The purest example on the board is the rook pawn with the wrong-coloured bishop. White has bishop and pawn against a bare king, an advantage that would win on any other file, and does not win: the light-squared bishop does not control h8, the square the black king settles on, and no manoeuvre dislodges him. The endgame tablebase returns a draw for every continuation, without exception.

What makes the position instructive is not the result, known for a century, but the mechanism: the defender has nothing to calculate. He occupies a square, the opponent cannot remove him from it, and a waiting move is always available. Recognising this at the board is worth more than any memorised variation — it is the difference between resigning and playing the ten moves that were left.

The rook-pawn corner, and the tempo that decides it

The count is short and it is a count of tempi: while the stronger king stands on g6, the squares g7 and h7 belong to him, and the defending king can only enter the corner through g8. If he is on f8 and to move, he enters; if he has to wait one move, h6-h7 shuts the door and the pawn queens. No bishop resource changes this, because the bishop takes no part — the race is between the kings.

The mistake seen in club games is the opposite of what you would expect. The defender, knowing that «wrong bishop with a rook pawn is a draw», stops running with the king and attends to something else — and arrives at the corner one move late. The tablebase draws the boundary precisely: black king on f8 to move, draw; black king on e7, loss; black king on d7, loss. There is no grey zone.

The other half of the trap catches the side that is winning. With the black king already on h8, the advance 1.h7 is immediate stalemate, and the natural developing move 1.Bd5 is stalemate too — the bishop starts covering g8 and takes the black king's last square away. Two moves that look like progress, both signing the draw on the spot.

Rook and pawn against the queen

The mechanism differs from the previous one. There is no unassailable corner here: there is a piece protected by a pawn about to promote, and a queen that cannot take it without giving the material back. If the queen takes the rook, the pawn queens; if she comes closer, the rook checks along the rank and returns to the block. The stronger king, meanwhile, cannot approach, because every approach square is covered by the rook-and-pawn combination.

The line in the diagram shows the pattern in three moves: 1.Qe2 Rf4+ 2.Kc5 Rf5+ 3.Kb6 Rf3 — the rook leaves with check, forces the white king to walk, and comes back to f3. This is the waiting move the first section's definition speaks of: while it exists, the position is shut. The tablebase confirms the draw in each of the six positions of that sequence.

This is the fortress that shows up most often in fast time controls, and also the one most often thrown away through ignorance: seeing queen against rook, the defender hands over the piece out of habit or tries desperate checks. The practical takeaway is simple — with a bishop or rook pawn on the seventh and the rook in front of it, look for the block instead of the counterattack.

Opposite-coloured bishops: counting files

This is the most common unbalanced endgame in club practice and the worst played, because counting pawns fails here completely. Two extra pawns decide any other endgame; here they do not, because the defending bishop can never be traded or driven off the squares it controls — the enemy bishop simply cannot see them. The stronger side has two attacking pieces and two fronts to cover at once.

The drawing diagram contains the whole mechanism in two black pieces: the king on b6 leans on the b5 pawn and the bishop on c7 occupies the b8-h2 diagonal, which runs through d6. While it stands anywhere on that diagonal the d5 pawn does not advance, and the bishop has waiting squares to spare. White can shuffle his own bishop indefinitely without producing anything, and the tablebase confirms the draw at every position in the sequence.

The same position with the d5 pawn moved to f5 changes the result. Now the squares to cover are b6 and f6, which do not share a diagonal, and the black king is tied to the queenside passer: the white king walks to the other wing and escorts the f-pawn, as in 1.Bc6 Bf4 2.Kd5 Bd2 3.Ke6. The boundary is exact: b5 and d5 draw, b5 and e5 still draw, b5 and f5 win.

• The defending bishop must match the colour of the pawns' advance squares, not the colour of the pawns.
• The king blocks the more dangerous passer; the bishop handles the other from a distance, on the right diagonal.
• Up to two files between the passers, the blockade holds; at three, it breaks.
• Material down with opposite bishops, every exchange of heavy pieces brings the draw closer.
• Before entering the endgame, check whether one of the enemy passers can be traded off — that is what shortens the distance.

When the fortress is not there

The line is short and worth memorising as a counterexample: 1.Bf5 Kg8 2.Kf6 Kh8 3.g6 Kg8 4.h7+ Kh8 5.g7#. Note that the extra pawn attacks nothing — it merely gives White something to do while the black king shuttles between g8 and h8. That is sufficient. Stalemate, which was the defence's resource, disappears the moment the stronger side can mark time.

The same principle explains the fortresses that collapse in long middlegames. Locked queenside pawns look irrelevant until a pawn break opens a second front; from then on the defender must cover two points with one king, and the barricade is walked around. Before trusting a fortress, count the pawn moves your opponent still has — if there is one, the fortress is temporary.

The third way to break a fortress is to sacrifice material to remove the piece holding it, and it is the one club players find hardest to see: giving up rook for bishop, for instance, to eliminate the guardian of the key square. If you are the stronger side and the position looks shut, look first for the waiting move; if there is none, look for the exchange that removes the guard.

The evaluation bar does not see fortresses

The reason is structural, and it reappears in other analyses. An engine evaluates by counting material and activity and corrects that count with search; a fortress is precisely the position where the search finds nothing, but also finds nothing bad — the static evaluation, a piece up, stays positive. The deeper the engine goes the closer to zero it drifts, and even then the bar can go on lying.

The consequence is direct for anyone reviewing games with automatic analysis. A drop to −0.8 in the endgame does not necessarily mean you erred: it may mean you entered a position the tablebase calls drawn and the engine does not. The remedy is to check the exact result once few pieces remain — tablebases cover up to seven pieces, and they hold no evaluations, only results.

This is not an argument against using the engine, but against reading the bar as a verdict. In open endgames Stockfish's evaluation is the best guide available; in locked ones it is an opinion about material. Telling the two cases apart is, in practice, the content of this guide — and it is also what separates the player who resigns a drawn position from the one who plays the ten moves that remained.

Frequently asked questions

Openings mentioned in this guide

• French Defense (C00)
• Caro-Kann Defense (B10)
• Ruy Lopez (Spanish Opening) (C60)