What is the first known example of published expert chess analysis that was disproven by a computer search, with the winning player completely opposite? That is to say, a widely known published analysis by a chess grandmaster of a position with the conclusion that it was winning for one player, but later a computer search proved that actually it was winning for the other player. This comes close, but I am curious to see examples with win-to-lose reversal of opinion. To clarify, I am looking for the first time we discovered that the previously accepted analysis was wrong, and our discovery arose from a computer search.

  • Are you interested in the oldest assessment that has been proven to be false by a computer (whenever the comp's evaluation has been revealed), or in the earliest time a comp evaluation reversed a preivously published analysis (even if that analysis was recent) ? What I mean is : A chess expert published an analysis showing that some position is winning in year X. A posterior analysis with computer refutes that evaluation and proves the position is losing in year Y. Are you looking for the earliest X or the earliest Y ?
    – Evargalo
    Commented Aug 30, 2019 at 10:06
  • 1
    @Evargalo: Sorry for the ambiguity. I meant "the earliest time a computer evaluation reversed a preivously published analysis". The idea is that it would be the first time that we are aware of a serious mistake that may not have been revealed to us without the computer analysis. I've edited my question to clarify that, thanks!
    – user21820
    Commented Aug 30, 2019 at 10:40

2 Answers 2


In the early eighties, the evaluation of the two bishops vs knight endgame was changed thanks to computer-assisted analysis.

Previously, that endgame was thought to be a draw, but now we know the bishops can force a win.

I don't know if this is the earliest instance, but it's definiely a quite old one

  • Can you add a source ? I thought the reassessment of KBBKN happened in the early nineties, but I might be wrong.
    – Evargalo
    Commented Aug 30, 2019 at 12:20
  • No. I can't add a source. I'm talking from the top of my head. I think I read it in a chess endgame book some years ago
    – David
    Commented Aug 30, 2019 at 12:33
  • 1
    It sure is mentionned in Alain Villeneuve's Les Finales. I just don't know since when ! The first edition was in 1982, but I only have had the ~1997 in the hands.
    – Evargalo
    Commented Aug 30, 2019 at 12:37
  • I plugged the endgame into tablebase (with the pieces all relatively centered) and it says White has a forced win but in over 50 moves (usually 75-100ish). So technically this means the endgame is a theoretical draw. Commented Aug 31, 2019 at 6:38
  • I thought I was quite explicit in my question asking for a win-to-lose reversal of opinion due to computer analysis.
    – user21820
    Commented Sep 2, 2019 at 16:08

Partial answers because this is an intrinsically difficult question, especially to prove it was the "first".

Endgame tablebases have changed the evaluations of some endgames. E.g.,

Queen versus two bishops. This was thought to be a draw due to the existence of a drawing fortress position, but the queen can win most of the time by preventing the bishops from getting to the fortress. However, it can take up to 71 moves to force a win (Nunn 2002:290ff).

The question also asks about grandmaster analysis being incorrect, with the actual result being the opposite. This certainly has happened before simply because grandmasters are fallible humans. Example by Tim Krabbe:

3r2k1/4qpbp/Q4P2/8/2N5/4p3/PP4PP/5RK1 w - - 0 1

Black to move. With the powerful passed pawn on e3, Black is clearly winning. 26...e2 27. fxe7 Bd4+ and White Resigns.

This game was published as a brilliancy, and it was months before someone spotted the fatal flaw in the combination: the position after 28...e2?? is actually winning for White. I won't spoil it here, but if you're interested you can find the solution on Tim Krabbe's website. Of course, in today's world, errors like this will never get published because an engine will spot the winning move in a fraction of a second.

  • 1
    Your example by Tim Krabbe would count if it had not been discovered until a computer analysis. But Tim's website says "a reader of the Observer chess column, which was one of the many where this combination was published, asked what happens if White plays 28.Ne3.". So, this doesn't count for the purposes of my question.
    – user21820
    Commented Sep 2, 2019 at 16:17
  • Yeah, I know. Same goes for the first bullet point, which didn't flip the evaluation from win to loss or vice versa. That's why I said partial answer.
    – Allure
    Commented Sep 2, 2019 at 19:40
  • @user21820 PS, maybe edit out the winning move for the next reader =)
    – Allure
    Commented Sep 3, 2019 at 3:32
  • Sorry, editing can only be done for the first 5 min. And thanks for your answer anyway!
    – user21820
    Commented Sep 3, 2019 at 10:45
  • @user21820 delete the comment and repost the relevant parts!
    – Allure
    Commented Sep 3, 2019 at 10:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.