Berlin, 1852, probably an informal game. Adolf Anderssen (White) v. Jean Dufresne (Black). Evans Gambit. 1–0, 24 moves as usually published.

Document 1. What we know, and how

Every move below is the standard published score, replayed by code before this page went up: each move is legal, each check is a check, the double check is a double check, and the last move is mate. The ClassicsForge bank on this site holds the score and is the authority if any copy disagrees.

Two cautions. When the game was first printed, the magazine did not give the final moves; it said only that White mates in four, which fits the position after 20.Rxe7+ Nxe7; announcing a forced mate instead of playing it out was ordinary practice. The four mating moves are the forced line everyone has accepted since, not a transcript. Second, the name came later. Wilhelm Steinitz, annotating the game after Anderssen's death in 1879, wrote of 19.Rad1 that it was "an evergreen in the laurel crown of the departed chess hero."

Document 2. The score, with notes

1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. b4. The Evans Gambit. White offers the b-pawn to gain time for c3 and d4.

4... Bxb4 5. c3 Ba5 6. d4 exd4 7. O-O. White is two pawns down and has castled; Black's king will stay on e8 all game.

7... d3. An odd choice: the pawn is advanced, a long way from support, and will be lost. It spends a tempo that Black's undeveloped kingside could have used.

8. Qb3 Qf6 9. e5 Qg6 10. Re1 Nge7 11. Ba3. The a3 bishop takes up the long diagonal toward e7 and will not move again until the final move.

11... b5 12. Qxb5 Rb8 13. Qa4 Bb6 14. Nbd2 Bb7 15. Ne4 Qf5 16. Bxd3 Qh5. Material is level: White has recovered both gambit pawns. Development is not. White has every piece in play except the a1 rook; Black's king is uncastled in the centre and the h8 rook has not moved.

17. Nf6+! A sacrifice and a fork: the knight checks the king and attacks the queen on h5. Black must take. 17... gxf6 18. exf6. The kingside pawn cover is gone and a white pawn stands on f6, next to e7 and g7. 18... Rg8. Black uses the open g-file.

19. Rad1! Steinitz's "evergreen". It brings the last white piece into play, behind the d3 bishop on the d-file. It also ignores the threat.

19... Qxf3? Black takes a knight and threatens 20...Qxg2#. This is the decisive mistake. Later analysis, notably Lipke's in 1898 and modern work since, found 19...Rg4 and other moves that may hold, possibly to a draw. The finish is forced only after Qxf3.

20. Rxe7+! Nxe7. The first of five consecutive checks. The engine check confirms that from here every white move gives check, so Black never gets time to play Qxg2.

21. Qxd7+!! Kxd7. Attraction: the queen sacrifice draws the king onto d7, the one square on the d-file in front of the d3 bishop and the d1 rook. (21...Kf8 was also legal.)

22. Bf5+. Double check, from the bishop on f5 and the uncovered rook on d1. Against double check the king must move. 22... Ke8. If 22...Kc6, then 23.Bd7# at once.

23. Bd7+ Kf8. 23...Kd8 also loses to mate next move.

24. Bxe7#. The a3 bishop's first move since move eleven. The f6 pawn covers g7 and guards e7, the d7 bishop covers e8, and g8 and f7 hold Black's own rook and pawn.

Document 3. Forcing moves

The finish illustrates the strongest principle in calculation: checks, captures and threats, in that order. When your opponent has a mating threat, you may only play moves that force a reply, because any quiet move allows the mate. A sequence of checks has a further property: it shrinks the tree of possibilities to almost nothing, which is what makes a five-move combination calculable at all. Count the legal replies at each step from move 20: three, two, two, two. A human can see that far precisely because the moves are forcing.

Document 4. Fork, attraction, discovery

Three tactical ideas carry the game, and they are worth naming exactly.

• Fork (17.Nf6+): one piece attacks two targets. Here the point is not to win the queen but to force gxf6 and destroy the pawn cover.
• Attraction (21.Qxd7+): a sacrifice that forces a piece, here the king, onto a particular square. It is the mirror image of deflection, which forces a piece off a square.
• Discovered and double check (22.Bf5+): a piece moves out of a line and uncovers an attack by the piece behind it, here while giving check itself.

Each one prepares the next. A combination is usually several small ideas in a row, each of which you already know, in an order you have not seen before. That is encouraging: you do not need new knowledge to find a combination like this one, only the habit of asking, at each step, which of the patterns you already know would work if one more piece were out of the way.

Document 5. Counting the replies

Here is the forcing line with the number of legal black replies after each white move, as the engine counts them: after 20.Rxe7+, three (Kf8, Kd8, Nxe7); after 21.Qxd7+, two (Kf8, Kxd7); after 22.Bf5+, two (Ke8, Kc6); after 23.Bd7+, two (Kf8, Kd8). Every branch ends in mate within the four moves except the ones Black did not choose at move 20 and 21, which need their own analysis.

Document 6. The objection

The game is often presented as a forced win from move 19. It is not. The correct verdict, as far as anyone can tell, is that 19.Rad1 set a problem that Black failed to solve, and that a better defence existed. That does not reduce the game. It changes what it teaches. The lesson is not "this attack was unstoppable". It is "when you have a forcing sequence, calculate it to the end; and when you are defending, look for the move that does not lose to it, not the move that grabs the most".

Document 7. The gambit, seventy years on

The Evans Gambit was invented by a Welsh sea captain, William Evans, in the 1820s, and for a generation it was among the most feared openings in chess. White gives a pawn and gets a fast centre and open lines. In this game Anderssen recovers both pawns by move sixteen and keeps the lead in development, which is the ideal outcome for the gambiteer.

The gambit later fell out of fashion as defensive technique improved; players found ways to return the pawn at the right moment and neutralise White's initiative. It still appears at the top level from time to time as a surprise weapon. The Evergreen Game shows the gambit at its best: not because the pawn sacrifice wins by force, but because Black spends the time it buys on queen moves instead of on castling.

Document 8. Reading a famous verdict

For forty-odd years after it was played, the standard view was that White's attack from move 19 was unstoppable. That view was repeated in books, and the repetition made it feel settled. Then Paul Lipke published analysis in 1898 showing that 19...Rg4 gave Black real drawing chances, and later analysts have proposed further defences.

It is worth asking how a mistaken verdict survives so long. Three reasons are typical. The original annotators were also the people who most admired the winner, which made them less likely to look for holes. The finish is so pleasing that readers want it to be forced. And checking a defence in a sharp position by hand is very hard work, so most readers simply trust the annotation. Engines have since removed the third reason. The first two are still with us, in chess and out of it.

Document 9. The two jobs

Look at the black pieces around the king just before 20.Rxe7+. The e7 knight blocks the e-file and also guards the king's flank. The c6 knight blocks the a4–e8 diagonal and also guards d8 and e7. The d7 pawn blocks the same diagonal. Every one of them is both a wall and a guard.

A piece doing two jobs is a target, because one forcing move can make it choose. The rook sacrifice on e7 makes the c6 knight recapture, which empties c6 and opens the queen's diagonal to d7. The queen sacrifice on d7 removes the last piece on that diagonal and pulls the king onto the d-file. Each sacrifice works by asking a defender to do one of its jobs at the expense of the other. That is a more general way of looking for combinations than memorising patterns: list each defender's jobs, and look for a move that forces it to drop one.

Document 10. Your position

Set up the position after 19.Rad1. You are Black. List every candidate move. For each, write down White's most forcing reply. Then compare your list with 19...Rg4. What did the defender need to see?

Then do one more exercise. Start from the position after 20...Nxe7 and write the mate in four as a tree: at each white move, list every legal black reply, and the white move that answers it. How many branches are there in total? That number is why a forcing line can be calculated in your head. Now try the same exercise from the position after 19.Rad1 with Black to move, listing Black's candidate moves and White's best reply to each. You will find the tree much wider, and it is in that wider tree that the drawing defences hide. Forcing lines are easy to see because they are narrow; defences are hard to find because the defender's tree is wide.