Game 1 is a French Defence Exchange Variation. White, outgraded, grabs a pawn - and holds it. Black overpresses.

A difference of X ELO points translates into an expected score of approximately (50+X/8) per cent for the stronger side. So a grading difference of 200 translates into 75%:25% (e.g 1 win 1 draw in two games, or 3 wins 1 loss in four games). So upsets happen - but not all that often.

In Game 2 White is outgraded by 188 ELO points but is not in the least overawed: he takes the fight to his opponent right from the start (accelerated Panov attack) and reaches a difficult N ending a Pawn up, and sacrifices it back to convert the full point.

Game 3 is another French Defence Exchange Variation. White is playing a GM. He tries to take his opponent out of the books and grabs material. By move 14 his position is difficult, by moves 17-22 it looks lost - then Black has an oversight.

Game 4 is a mating attack by an outgraded Black:

click for larger view

(Black to play, 26...?)

Game 5 shows a logical build-up by the White underdog exploiting Black's ambitious play. The combination proper starts with White to play at move 24:

click for larger view

In Game 6 Botvinnik, fourteen, beats the World Champion with Black (in a simultaneous). Playing through the game, one asks - where exactly does White go wrong? - he plays natural moves and gets completely smashed (OK, 9.Bxf6 Nxf6 10. O-O-O was decidedly optimistic).

In game 7 the game is heading peacefully for a draw

click for larger view

when the grandmaster, Black, inexplicably liquidates into a should-be-drawn ending with 51...Qxg4 52.Nxg4 e4 - and then lost it. There was a theoretical draw by the fifty-move rule but the ending is tricky to defend.