leonelcm> If white do the right and also black, then is a draw. Any tiny little mistake in one side is win for the other.

Err, no, it's a tablebase win--although somewhat complicated for humans. See normajeanyates' message and my follow-up diagram. :)

 edited above post - reposting also

 Slight correction - *white's* play (flying rook defence) is somewhat complicated for humans. *black's* strategy (of winning against flying rook defence) - though not in the tablebase way i showed of course! - was already known to humans by 1992. (K+Q v K+R was one of the first tablebases made - by Ken Thompson of course - thats how the frd was discovered - but human strategies to beat it were soon found). By now a majority of IMs - and definitely almost all GMs - know it. All FMs could master winning Q v R against tablebases if they spent a week learning it and practicing it (hell i plan to learn it - it is describable in 1 page --- search rec.games.chess for Roger Poehlman's 1992 post and you can find it.) Masters planning to play official/demo game against a computer learn it as part of preparation if they havent learnt it already.

 It is much simpler than the troitsky (two N's v P with P behind troitsky line) - in the 1940s one USSR GM reached it twice and failed to win both times - USSR chess team captain - the legendary Mikhail Botvinnik - suspended him from further official play until he had leaned it - in a fortnight he did - botvinnik tested him and reinstated him.

[tablebases have found that in some cases the troitsky is a win even with the pawn ahead of the troitsky line, but it is beyond humans to win those *ultra-troisky* positions against computers.]