Endgames — A Small, Ordered Curriculum 慎終如始

There are perhaps twenty-five positions worth an improver's time between 600 and 1800. Not five hundred. This page lists them in order, describes each one in words, and gives you the drill. Everything that is usually padded into an endgame book has been left out on purpose.

The one-paragraph version

Learn the basic mates until they are boring. Learn king-and-pawn properly — it is the only endgame that is genuinely solved by rules you can memorise. Learn Lucena and Philidor at around 1200. After that, stop collecting positions and start working on practical technique: activate the king, trade pieces rather than pawns when ahead, put the rook behind the passed pawn, and do not hurry. Most endgames you actually lose will be lost to one of those four, not to a theoretical position you had never seen. That last sentence is a judgement, not a measurement — see the red card below.

First, the contested part

The Ladder puts endgames at Rung 2, from about 900. That is earlier than most advice, and the argument deserves to be laid out properly rather than asserted.

The case for early

The only phase-by-phase blunder measurement we can find published anywhere — ~162,000 Lichess rapid games analysed with Stockfish 17, blunder defined as a move losing 300 centipawns or more — finds the endgame phase is where amateurs are worst and where they improve least:

Phase	400–600	1500–1700
Opening	19.6%	7.1%
Middlegame	43.2%	30.9%
Endgame	45.9%	39.0%

Opening blunders fall by about two thirds across that range. Endgame blunders fall by about a seventh. Whatever players are learning as they climb, it is not this.

Those band labels are chess.com-equivalent: the games are Lichess games, mapped across by a community conversion table. This site's own Sources page notes that the published conversion tables are irreconcilable, so read the bands as approximate.

The case for later

A statistical study of the Lichess September 2017 database — 12.6M games, 2.74M of them classical — found that only 14% of those classical games reached a technical endgame at all under its definition (three pawns a side or fewer, at most two pieces a side, no crushing material edge, at least forty moves each).

Nate Solon puts the sceptical case plainly: "most games between beginners do not reach the endgame at all," and "I honestly can't recall a single game where failing to know one of those endgames cost me even half a point" — those being the contents of 100 Endgames You Must Know. He names the genuinely common ones in passing (the basic mates, king and pawn, Philidor) but pointedly declines to prescribe a list at all: his conclusion is "study what you like while maintaining some balance between different parts of the game."

Both numbers have real problems, and nobody arguing about this mentions either

The blunder figure flatters the endgame. Centipawn loss is not comparable across phases. In a simplified position a single king move flips the theoretical result from draw to loss, and the engine registers that as an enormous swing. In a murky middlegame an error of similar magnitude may cost 150 centipawns and not count. The direction of the finding is probably right — endgame skill is not improving with rating — but the size of the gap is partly an artefact of the metric. Single unreplicated source, small commercial site, stated methodology.

The 14% figure is strict and old. It is one month of one site, classical only, in 2017, and its material criteria exclude a lot of positions an improver would call an endgame: four pawns a side disqualifies the game, as does more than thirteen points of material or more than two non-pawn pieces a side. Loosen the definition and the number rises. Nobody has published the sensitivity analysis. The source is also internally inconsistent on the rook figure — it reports 67% of endgames containing a rook and, a paragraph later, that 37% do not.

And the question underneath both is unmeasured. No study has ever tested whether studying endgames raises anyone's rating, at any level, against a matched control. Not one. Everything on this page — including the ordering — is engineering judgement built on descriptive data and coach consensus, and you should hold it that loosely.

Capablanca is the usual authority for endgames-first, via "in order to improve your game, you must study the endgame before everything else." We could not find that sentence in the Project Gutenberg text of Chess Fundamentals, and no quotation site gives a page reference — treat the wording as unsourced. What is verifiable is structural and arguably stronger: Chapter I of Chess Fundamentals (1921) is titled "First Principles: Endings, Middle-game and Openings," and it opens with simple mates and pawn endings before it mentions an opening.

This site's position

The curriculum below is small enough that the argument stops mattering. Tier 0 and Tier 1 together are about ninety minutes of drilling plus a spaced revisit queue. Even on the sceptic's numbers that is worth it, because the positions are cheap, they never expire, and unlike an opening line nobody can take you out of them. Where the two camps genuinely diverge is at Tier 3 and above — and there this page agrees with the sceptics. If you are under 1500, do not buy a comprehensive endgame reference — Dvoretsky, Müller and Lamprecht — and try to work through it.

Tier 0 — under about 900 · the mates, and not throwing them away

One job here: convert an overwhelming material advantage without stalemating. That is the whole tier.

Position	The key idea	The drill
Two rooks vs lone king The ladder or staircase mate.	Cut with one rook, check with the other, walk the king to the edge. Your own king is not needed.Two rooks vs lone king (representative) · White to move, mate in 6	Ten mates in a row, each in under ten moves.
King + queen vs king	Shrink the box with the queen, keeping her a knight's move from the enemy king, then bring your own king up to deliver. With the queen's side to move, this is mate in at most ten moves from any position.K+Q vs K · a worst case: mate in 10	Twenty mates from random starts. Any stalemate resets the counter to zero.
King + rook vs king	Cut the king off on a rank or file, take the opposition with your king, then squeeze one rank at a time. Slower and far more error-prone than the queen — at most sixteen moves with best play and the rook's side to move, and beginners routinely take fifty.K+R vs K · a worst case: mate in 16	Twenty mates from random starts, against an engine that defends properly.
The stalemate you will actually cause	Black king a8; white queen b6; white king on any other legal square — it is stalemate from every one of them. Black is not in check and has no legal move. That is a draw, and it is how most beginners lose a won game.The stalemate you will actually cause · Black to move and has no move	Set it up once and look at it. Then adopt the rule: with a lone enemy king, never place your queen adjacent to it unless your own king is defending her.

Bishop-and-knight mate is deliberately absent, here and from Silman, who wrote that he "heretically decided not to include Bishop and Knight vs. Lone King because it's far from easy to master, and it occurs very rarely in over-the-board play." That is the right call for a time-poor adult. Learn it at 2000 if you want it as an ornament.

Tier 1 — about 900 to 1200 · king and pawn

This is the highest-value tier on the page, because king-and-pawn endings are the one part of chess that is genuinely reducible to rules. You are not building intuition here. You are memorising four facts and then executing them.

The rule of the square Count the squares the pawn needs to promote. Draw a square of that side length from the pawn towards the promotion rank. If the defending king is inside that square, or can step inside on its move, it catches the pawn. If not, the pawn runs. Adjust for the double step: a pawn on its starting rank counts as if it were one rank further advanced.Rule of the square (representative) · White to move wins; Black to move ...Ke6 draws Drill: twenty positions, answer in under three seconds each.

Key squares For a pawn on the second, third or fourth rank, the key squares are the three squares two ranks in front of it — for a pawn on e4, those are d6, e6 and f6. For a pawn on the fifth or sixth rank there are six: the three squares one rank ahead and the three two ranks ahead. Get your king onto a key square and the pawn queens whoever is to move.Key squares · king on d6 (a key square for the e4 pawn): White wins either way Drill: name the key squares for ten random pawns.

The opposition, and the one rule that overrides it White Ke6, pawn e5, Black Ke8: White wins with either side to move — a king on the sixth rank in front of its own pawn wins regardless of the opposition, provided it is not a rook pawn.King on the sixth in front of its pawn · White wins with either side to move Contrast White Ke5, pawn e4, Black Ke7: with White to move Black holds the opposition and draws;Opposition · White to move: Black holds the opposition and draws with Black to move White wins. Same material, same shape, opposite results.The same position, Black to move · White wins Drill: play both positions out from both sides against an engine until you never get them wrong.

The rook pawn is different If the defending king reaches any square in front of a lone rook pawn it cannot be driven off the file, and the pawn never promotes. White Kh6, pawn h5, Black Kh8 is a draw.The rook pawn · drawn with either side to move The same rule makes the notorious wrong-bishop case: an a-pawn with a bishop that does not control a8 wins nothing if the defending king sits on a8, a7 or b8. It shuffles there for ever, and stalemate is often the finish.Wrong bishop with an a-pawn (representative) · drawn, Black just shuffles a8/b8

All four verdicts above — e5/e6 wins either way, e4/e5 is a draw only with White to move, h5/h6 is a draw either way — were checked against the Lichess 7-piece tablebase rather than taken from a book. Free drill sources for all four: the Lichess Practice section's Pawn Endgames chapters include Key Squares and Opposition, and both play out against a computer that defends correctly. See Tools.

Tier 2 — about 1200 to 1500 · two rook positions and two rules

Rook endings are where the study budget belongs. In that Lichess sample, 67% of games that reached a technical endgame contained rooks, and the two commonest configurations of all were rook-and-pawns against rook-and-pawns (14.8%) and rook-and-one-pawn against rook-and-pawns (14.1%). Nothing else is close to a rook ending individually — though pure pawn endings, taken together, are the next block at roughly a fifth of the sample, which is the honest argument for Tier 1 being the tier you must not skip.

The Lucena position — how to win

White king b8, white pawn b7, white rook c1; black king d8, black rook a2. White to move. The pawn is one square from queening and White's own king is in the way.Lucena · White to move and wins The technique is building a bridge.

1.Rd1+ Ke7 2.Rd4! Ra1 3.Kc7 Rc1+ 4.Kb6 Rb1+ 5.Kc6 Rc1+ 6.Kb5 Rb1+ 7.Rb4!

Move 2 is the whole idea and it looks pointless at the time. The rook goes to the fourth rank so that later, when the checks come, it can drop in on b4 defended by its own king on b5. Black must either take it — and after the recapture the pawn queens — or stop checking.Lucena · the bridge after 7.Rb4!

The line above was replayed for legality and checked move by move against the Lichess 7-piece tablebase: the starting position is a win, and every White move in it holds the win. Named after Luis Ramírez de Lucena, and the attribution is wrong: the position does not appear in his 1497 book. The earliest known discussion is Salvio's Il Puttino, 1634, where Salvio credits it to Scipione Genovino. The mistake seems to come from the sixth edition of the Handbuch des Schachspiels (1880), whose editor Constantin Schwede attributed it to "Lucena 96."

The Philidor position — how to draw

Described in words, because the defender's setup is the thing to remember and the attacker's king can stand anywhere. Say White has a pawn on e5 and you are Black. Your king is in front of the pawn — on e8 — and your rook sits anywhere on the sixth rank, say a6.Philidor (attacker's setup representative) · Black draws

While your rook holds the sixth rank, the white king cannot step onto d6, e6 or f6. Without its king in front, the pawn cannot advance safely.
So White's only try is to push the pawn to e6 himself. The instant that happens, your rook drops to the first rank, on a file where it is safe — and checks from behind.Philidor · after e5-e6 the rook drops to a1 and checks from behind — still drawn Usually a1. Not blindly a1: if White’s rook happens to sit on the first rank too, …Ra1 just hangs.
The white king now has nowhere to hide, because its own pawn no longer stands in front of it as a shield. The checks never stop. Draw.

Philidor analysed this in 1777. It is the most valuable defensive position in chess for an improver, because rook endings with one side a pawn down are, on the Lichess numbers above, the single likeliest shape you will find yourself defending.

The short side, the long side, and the rook behind the pawn

If your king is driven off the pawn's file, go to the short side. Put your king on the side with fewer files between the pawn and the edge, and check from the long side with your rook. Your rook needs running room: a rook one or two files from the pawn gets chased off by the enemy king, and the defence needs at least three clear files between rook and pawn — equivalently, the rook four files over. Sources state it both ways and it looks like a disagreement; it is not, it is the same requirement counted inclusively or exclusively. Getting your king out of the rook's way is the whole point of choosing the short side.Short side defence (representative) · king short side, rook four files away — Black draws
Rooks belong behind passed pawns — behind your own to push them, behind the opponent's to hold them. Tarrasch stated it in his book on St Petersburg 1914 and called it the most important rule in complicated rook endings in The Game of Chess (1938). He also, to his credit, added: "Always put the rook behind the pawn... Except when it is incorrect to do so." Blocked pawns, pawns still on the fourth rank and rook's pawns are all real exceptions.Rooks and the passed pawn (representative) · White's rook behind its own pawn, Black's in front

Tier 3 — about 1500 to 1800 · the ideas, not the positions

The material changes character here. Below 1500 you are learning results; above it you are learning judgements, and there are fewer of them than the books suggest.

Rook activity beats material

A rook behind a passed pawn on the seventh, or raking an open rank, is routinely worth a pawn — and passive rook defence loses endings that active defence draws. This is the most repeated heuristic in rook endings and it is a heuristic: nobody has measured the exchange rate. Treat "an active rook is worth a pawn" as a bias to adopt, not a valuation.

The Vancura position

White king d4, rook a8, pawn a6; black king g7, black rook f6. Black draws. The rook attacks the pawn from the side along the sixth rank, which ties White's rook to defending it permanently, and the black king stays on g7 or h7 so that White can never sacrifice the pawn and skewer king and rook on the seventh. This is the standard save when you are a rook-pawn down; the tablebase confirms the diagrammed position is drawn.Vancura · Black draws Provenance is genuinely unresolved: Vančura died in 1921, and Edward Winter reports being unable to exhibit the position in a 1924 source, though mid-1920s issues of Časopis Československých šachistů reproduced many of his compositions posthumously, noting also an unexplained link to a Tarrasch analysis of 1908. "Published posthumously" is the usual story rather than a documented fact.

Opposite-coloured bishops

With opposite bishops and nothing else, one pawn up is normally a draw and two or three extra pawns may not be enough. With bishop and two pawns against bishop, roughly half of endings are drawn, against over 90% wins with same-coloured bishops.Opposite bishops, one pawn up (representative) · drawn If you are winning, this is the ending to avoid entering; if you are losing, it is the one to steer for. Winning needs passed pawns widely separated — one file apart is usually a draw, three files apart usually wins.Opposite bishops, passers three files apart (representative) · White wins

Good knight, bad bishop

The knight is better with pawns on one wing and a blocked structure; the bishop is better with pawns on both wings and open diagonals. The practical version: before you trade into a minor-piece ending, ask which wing the pawns are on and whether the position is fixed.Good knight, bad bishop (representative) · fixed pawns, all on one wing What makes a bishop "bad" in the first place, and the two things to do about it, are on Strategy.

The outside passed pawn

A passed pawn far from the main pawn mass wins king-and-pawn endings by decoying the enemy king. Your king then eats the abandoned wing.The outside passed pawn (representative) · White wins Coaches treat this as the commonest winning mechanism in amateur pawn endings; nobody has counted, so take it as the strong consensus it is rather than a statistic. Even discounted, it is worth more study time than any single theoretical position above.

The breakthrough

Three connected pawns on a5, b5, c5 against a7, b7, c7, kings far away:The breakthrough · 1.b6! (kings placed far away — representative)

1.b6! axb6 2.c6! bxc6 3.a6

and the a-pawn needs just two more moves to queen — three clear moves ahead of Black’s fastest pawn. If instead 1...cxb6 2.a6! bxa6 3.c6 and the c-pawn does. Two pawns for a queen. Both lines were replayed for legality; the shape is rare, but it is free to recognise and unfindable if you have never seen it.

Triangulation and zugzwang. If you need the same position with the opponent to move, walk your king in a triangle — three moves to reach a square you could have reached in two — and hand the move over. Zugzwang is the engine of every king-and-pawn ending, and triangulation is how you manufacture it. Drill it in a simple pawn ending until you stop thinking of it as a trick.Zugzwang in a blocked pawn ending (representative) · White to move wins, Black to move draws

Tier 4 — 1800 and up · what actually comes next

Briefly, because at this point you no longer need a curriculum written for you. Rook and bishop against rook (a theoretical draw that is genuinely hard to hold, and the fifty-move rule is part of the difficulty). Queen against rook. Rook and two pawns against rook. Corresponding squares in blocked pawn endings. Queen endings, which are mostly about perpetual check and the far-advanced passer. This is where de la Villa's 100 Endgames You Must Know earns its title, and where Dvoretsky's Endgame Manual becomes usable rather than punitive.

The part that actually decides your games

Everything above is theoretical knowledge. Most of the endgames you lose are lost to technique, and technique is four habits.

Activate the king the moment the queens come off Not "eventually." That move. In the endgame the king is a fighting unit worth about a minor piece — Lasker put it at a knight plus a pawn — and the side whose king gets to the centre first is usually the side that wins the pawn race. That last clause is coaching lore, not a count. The habit is free, it costs no study, and most players below 1600 simply forget.

Ahead in material, trade pieces — not pawns And when behind, trade pawns and keep pieces. An extra piece matters more as the board empties; an extra pawn is worthless if you trade away all the pawns and reach a drawn ending. Heuristic, universally taught, no measurement behind it, and it is right often enough to be a default. The rest of the trade rules — cramped, more space, bad piece for good piece — are tabulated on Strategy.

Do not hurry The principle Shereshevsky made famous — chapter 12 of Endgame Strategy, Soviet original early 1980s, English edition 1985 — and the hardest one to obey. He gives it in quotation marks, as received Soviet endgame doctrine rather than his own coinage, and no earlier source for the phrase is firmly established. Before committing to the winning plan, improve every piece to its best square, put your pawns where you want them, and only then break. Rushing a conversion is the commonest way a winning ending becomes a draw at club level.

Create a second weakness The endgame case of the principle set out on Strategy: one weakness can be defended, two on opposite wings cannot, because the defender's king cannot be in both places. Noted here for completeness and then deliberately parked — Silman files it under Master, 2200+. It is not what is costing you points at 1400, and studying it before you can win Lucena cold is the temptation this whole site is built against.

How to train endgames

Endgames are the one area where the difference between reading and drilling is total, because the material is procedural. Reading an endgame book is the purest available form of 罔 — Principle Ⅲ, learning without practice — and it feels productive the entire time.

篤

Play it out against an engine, from both sides This is the fifth gate of Principle Ⅵ, and endgames are its clearest application. An engine defends perfectly and never gets bored, which is exactly what you need. Standard: win or hold the position three times in a row with no hints, then switch colours and do it again from the losing side. Knowing the theory of the Lucena position is not the same as winning it with a clock running.

溫

Put every drilled position in the spaced queue Principle Ⅰ. Same queue as your failed puzzles: roughly 7 days, then 3 weeks, then 2 months. Re-derive the method from the diagram, cold. Do not reread the solution — if the revisit teaches you nothing new, you reread rather than re-derived and it did not count. Endgame technique decays faster than tactical pattern recognition because you use it less often.

明

Interleave once you know a theme Ten Lucenas in a row teaches you to execute a method you were told to use. A mixed set of rook endings forces you to work out which method the position needs, which is the actual skill. Rohrer and Taylor's numbers on this are on the method page and they are brutal.

Tablebases: a checking tool, not a learning tool

Lichess has hosted the complete 7-piece Syzygy tablebases since 19 August 2018 — 18.4 TB covering all 423,836,835,667,331 legal 7-piece positions, free on the analysis board. It will tell you instantly whether the ending you just drew was actually winning.

Use it for exactly that and nothing else. A tablebase gives you the result and never the reason, and its lines are frequently unlearnable: the longest 7-piece win found runs 1,034 half-moves before the first winning capture. That is not a technique. Consult it after you have committed to your own assessment — same discipline as engine-off-first analysis in Principle Ⅱ, and for the same reason. Once you have seen the tablebase move you cannot reconstruct why you would have played something else.

The Silman approach, and the instruction most people ignore

If you buy one endgame book, buy Silman's Complete Endgame Course. Its structure is the reason: 528 pages divided into nine parts by rating band, and you are told to read only yours.

Part	USCF band	What is in it
One	unrated–999	Overkill mates, the staircase, the box, K+Q vs K, K+R vs K, Stalemate Alert
Two	1000–1199	What can and cannot mate a lone king, minor piece vs queen, understanding the king
Three	1200–1399	Beyond basic opposition, king and pawn endings, minor piece vs a lone pawn, rook vs a lone pawn
Four	1400–1599	Lucena and Philidor, king and pawn, minor piece endings, queen vs king and pawn
Five	1600–1799	More king and pawn, rook endings, minor piece endings
Six to Nine	1800+	Class A, Expert, Master — the last of which is where the Principle of Two Weaknesses lives — then a part of endgames for pure pleasure

Silman's own instruction is explicit: "if you have just learned to play, all you need to study is the section designed for beginners (Part One). After mastering the material there, put the book away and spend your time studying tactics and a few strategic concepts." Almost nobody does this. The book is bought, read straight through in a fortnight, half-remembered, and none of it is drilled. That is a worse outcome than never opening it, because it consumed the study hours and produced a feeling of competence — Principle Ⅳ, exactly.

Two calibration notes. Silman's bands are USCF; on this site's conversion figures those map roughly onto chess.com rapid at similar numbers in the middle of the range, drifting apart at the top. And his Part Four is a little later than this page's Tier 2 for Lucena and Philidor — he wants them at 1400, we want them at 1200. That is a genuine disagreement between a titled coach's teaching experience and this site's reading of the blunder data. He may well be right. De la Villa has also published a gentler companion to his famous book, 100 Basic Endgames You Must Know, pitched at club players — but it appeared in 2026 and has no track record yet, so this is a pointer, not a recommendation.

民之從事，常於幾成而敗之。慎終如始，則無敗事。

Mín zhī cóng shì, cháng yú jī chéng ér bài zhī. Shèn zhōng rú shǐ, zé wú bài shì.

"People in their conduct of affairs are constantly ruining them when they are on the eve of success. If they were careful at the end, as they should be at the beginning, they would not so ruin them."

老子 Daodejing 64 · tr. Legge

A different school from the Confucian texts used elsewhere on this site, and flagged as such. The received text reads 慎終如始; the Guodian bamboo-slip recension has 慎終若始. Edition difference, not error. It is quoted here because it is the most exact description of a club endgame that exists in any language.

Your whole endgame budget, if you are under 1500

Ninety minutes once, then fifteen minutes a week. The ninety minutes buys Tier 0 and the four rules of Tier 1. The fifteen minutes a week is the spaced queue plus one played-out position. At around 1200 you add Lucena and Philidor, which is another hour. That is the entire investment — and pound for pound it is the most durable knowledge on this site, because unlike an opening repertoire nobody can prepare against it, and unlike a tactical pattern it will still be exactly true in thirty years.