Position : Win or Draw ?

[Robert Flesher]

Hello folks, some may have forgotten that Chiron has pawn blockage recognition. Chirons sees a dead draw!New game Line
3k4/1p3b2/1Pp3p1/2Pp1p2/3P1P2/8/4K1N1/8 w - - 0 1

Analysis by Chiron 2 64bit:

1.Kf2 Ke7 2.Ne1 Kf6 3.Nf3 Be8 4.Kg3 Kf7 5.Kh4 Kg7 6.Kg5 Bd7 7.Ne5 Be8 8.Nf3 Bd7 9.Ne5 Be8 10.Nf3
The position is equal: = (0.00) Depth: 84/107 00:03:38 852MN
(, 16.12.2019)

[peter]

Hello folks, some may have forgotten that Chiron has pawn blockage recognition. Chirons sees a dead draw!New game Line
3k4/1p3b2/1Pp3p1/2Pp1p2/3P1P2/8/4K1N1/8 w - - 0 1

Analysis by Chiron 2 64bit:

1.Kf2 Ke7 2.Ne1 Kf6 3.Nf3 Be8 4.Kg3 Kf7 5.Kh4 Kg7 6.Kg5 Bd7 7.Ne5 Be8 8.Nf3 Bd7 9.Ne5 Be8 10.Nf3
The position is equal: = (0.00) Depth: 84/107 00:03:38 852MN
(, 16.12.2019)

Thanks, Robert, good to remember other engines but SF and LC0 now and then too, Chiron's a good one also for sure, but all other engines (I tried) see the draw in the original postion at once too anyhow. Most of them have a little more then 0.00 started with empty hash. It was Sting only, that was dreaming about a winning eval in standalone- pondering, yet also only till you let it play out some moves on its own, and Marek's already after it:
viewtopic.php?p=822594#p822594

[d]3k4/1p3b2/1Pp3p1/2Pp1p2/3P1P2/8/4K1N1/8 w - - 0 1

Analysis by Stockfish 101219 64 POPCNT:

1.Kf2 Kd7 2.Nh4 Ke8 3.Nf3 Kf8 4.Kg3 Kg7 5.Ne5 Be8 6.Kh4 Kf6 7.Nd3 Bd7 8.Nf2 Kg7 9.Kg5 Be8 10.Nd1 Bd7 11.Nc3 Be8 12.Kh4 Bd7 13.Ne2 Be8 14.Ng1 Bf7 15.Kg5 Be8 16.Nf3 Bd7 17.Nh4 Be8 18.Ng2 Bd7 19.Kh4 Be6 20.Ne1 Bf7 21.Kh3 Be8 22.Nf3 Kf6 23.Kg3 Kf7 24.Ng5+ Ke7 25.Kh4 Bf7 26.Nh3 Kf6 27.Kg3 Kg7 28.Kf3 Kf6 29.Ng5 Bg8 30.Kg3 Bf7 31.Kh2 Kg7 32.Kg2 Kf6 33.Kg3 Kg7 34.Kf3 Kf8 35.Ke3 Ke7 36.Nf3 Kf6 37.Kd3 Be8
+/= (0.53) Depth: 67/74 00:00:34 524MN
...
1.Kf2 Kd7 2.Nh4 Ke8 3.Nf3 Kf8 4.Kg3 Kg7 5.Ne5 Be8 6.Kh4 Kf6 7.Nd3 Bd7 8.Nf2 Be8 9.Nh3 Bf7 10.Kg3 Kg7 11.Kf2 Kf6 12.Kf3 Bg8 13.Ke3 Kg7 14.Nf2 Kf6 15.Nd3 Bf7 16.Ne5 Be8 17.Nf3 Bd7 18.Kd3 Be8 19.Kc3 Bf7 20.Ng5 Bg8 21.Kc2 Ke7 22.Nf3 Bf7 23.Ne5 Be8 24.Kb1 Kf6 25.Nf3 Bd7 26.Ng5 Ke7 27.Nh7 Kf7 28.Kc2 Ke7 29.Kd3 Be8 30.Ng5 Bd7 31.Nf3 Kf6 32.Ke3 Be8 33.Ng5 Ke7 34.Kd2 Bd7 35.Kd3 Be8 36.Kc2 Kf6 37.Kc3 Ke7 38.Kb4 Bd7 39.Ka5 Be8 40.Nh3 Bd7 41.Ka4
+/= (0.53) Depth: 76/81 00:01:21 1849MN

And after some very quick Forward- Backward of such a first one output- line, 0.00 or at least eval in = range is kept in hash.

Crystal's fortress detection works well here too, started with empty hash:

3k4/1p3b2/1Pp3p1/2Pp1p2/3P1P2/8/4K1N1/8 w - - 0 1

Analysis by Crystal 011219 64 POPCNT:

1.Ne1 Ke7 2.Nf3 Kf6 3.Ng5 Be6 4.Kf3 Ke7
= (0.11) Depth: 7/13 00:00:16 70kN
1.Ne1 Ke7 2.Nf3 Kf6 3.Ke3 Be6 4.Ne5 g5 5.Kf3 Bg8
= (0.30) Depth: 8/13 00:00:16 79kN
1.Ne1 Ke7 2.Nf3 Be8 3.Kf2 Ke6 4.Kg2 Kf7 5.Ne5+ Kf6 6.Kf3 g5 7.fxg5+ Kxg5
= (0.00) Depth: 9/16 00:00:16 147kN
...
1.Ne1 Ke7 2.Nf3 Be8 3.Ne5 Ke6 4.Ke3 Kf6 5.Nd3 g5 6.fxg5+ Kxg5 7.Ne5 f4+ 8.Kf2 Kf5 9.Kf3 Bh5+ 10.Kf2
= (0.00) Depth: 47/47 00:07:10 9562MN

So Michael Byrne came up with an alternative position with black knight instead of bishop and an white e-pawn added.

[d]3k4/1p3n2/1Pp3p1/2Pp1p2/3P1P2/4P3/4K1N1/8 w - - 0 1

That being a safe fortress too, it's much harder for all engines I tried, to get the drawing eval in pondering or playing forward and or at least keep it in hash in backward analysing.

viewtopic.php?p=822537#p822537

[Ovyron]

Don't know, if you'll find more useless pawns to add, but the one you added this time, already isn't useless at all anymore of course, guess you know so, don't you?

I wasn't kidding, I still have problems understanding those fortresses, it'd probably help me to see when one exists at a glance without having to fire up an engine.

I now wonder if the position can be modified in a way where it's not trivial to know if the position is a mate or a draw.

[fishpov]

Thanks Marek

[Dann Corbit]

Hello folks, some may have forgotten that Chiron has pawn blockage recognition. Chirons sees a dead draw!New game Line
3k4/1p3b2/1Pp3p1/2Pp1p2/3P1P2/8/4K1N1/8 w - - 0 1

Analysis by Chiron 2 64bit:

1.Kf2 Ke7 2.Ne1 Kf6 3.Nf3 Be8 4.Kg3 Kf7 5.Kh4 Kg7 6.Kg5 Bd7 7.Ne5 Be8 8.Nf3 Bd7 9.Ne5 Be8 10.Nf3
The position is equal: = (0.00) Depth: 84/107 00:03:38 852MN
(, 16.12.2019)

Crafty had Jeremiah Pennery's blockade detection starting with Crafty 19.1 in 2002 (17 years ago):

Code: Select all

White(1): setboard 3k4/1p3b2/1Pp3p1/2Pp1p2/3P1P2/8/4K1N1/8 w - -
White(1): st 99
search time set to 99.00.
White(1): sd 99
search depth set to 99.
White(1): go
              clearing hash tables
              time surplus   0.00  time limit 1:39 (1:39)
              depth   time  score   variation (1)
starting thread 1
starting thread 2
starting thread 3
starting thread 4
starting thread 5
starting thread 6
starting thread 7
starting thread 8
starting thread 9
               16     0.93  -0.24   1. Nh4 Ke7 2. Nf3 Kf6 3. Ke1 Be8 4.
                                    Kf2 Kg7 5. Kg3 Kh6 6. Ng5 Kh5 7. Nf3
                                    Bd7 8. Ne5 Be6 9. Nxc6 <HT>
               16->   1.04  -0.24   1. Nh4 Ke7 2. Nf3 Kf6 3. Ke1 Be8 4.
                                    Kf2 Kg7 5. Kg3 Kh6 6. Ng5 Kh5 7. Nf3
                                    Bd7 8. Ne5 Be6 9. Nxc6 <HT>
               17     1.47  -0.24   1. Nh4 Ke7 2. Nf3 Kf6 3. Ke1 Be8 4.
                                    Kf2 Kg7 5. Kg3 Kh6 6. Ng5 Kh5 7. Nf3
                                    Bd7 8. Ne5 Be6 9. Kf3
               17->   1.65  -0.24   1. Nh4 Ke7 2. Nf3 Kf6 3. Ke1 Be8 4.
                                    Kf2 Kg7 5. Kg3 Kh6 6. Ng5 Kh5 7. Nf3
                                    Bd7 8. Ne5 Be6 9. Kf3
               18     2.45  -0.24   1. Nh4 Ke7 2. Nf3 Kf6 3. Ke1 Be8 4.
                                    Kf2 Kg7 5. Kg3 Kf7 6. Ne5+ Ke6 7. Nf3
                                    Kd7 8. Ne5+ Ke7 9. Nf3 Bd7 <HT>
               18->   3.79  -0.24   1. Nh4 Ke7 2. Nf3 Kf6 3. Ke1 Be8 4.
                                    Kf2 Kg7 5. Kg3 Kf7 6. Ne5+ Ke6 7. Nf3
                                    Kd7 8. Ne5+ Ke7 9. Nf3 Bd7 10. Kh4
                                    <HT> (s=3)
               19     6.40   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kf8 6. Kg5 Kg7 7. Ne5
                                    Bf7 8. Nf3 Be8 (s=2)
               19->   7.25   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kf8 6. Kg5 Kg7 7. Ne5
                                    Bf7 8. Nf3 Be8 (s=2)
               20     9.86   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kf8 6. Kg5 Kg7 7. Ne5
                                    Bf7 8. Nf3 Be8
               20->  11.34   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kf8 6. Kg5 Kg7 7. Ne5
                                    Bf7 8. Nf3 Be8 (s=2)
               21    15.45   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kf8 6. Kg5 Kg7 7. Ne5
                                    Bf7 8. Nf3 Be8
               21->  18.33   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kf8 6. Kg5 Kg7 7. Ne5
                                    Bf7 8. Nf3 Be8 (s=2)
               22    26.20   0.01   1. Nh4 Ke7 2. Nf3 Kf6 3. Kf2 Be8 4.
                                    Kg3 Kf7 5. Kh4 Kg7 6. Kg5 Bd7 7. Ne5
                                    Be8 8. Nf3 <HT>

[peter]

I now wonder if the position can be modified in a way where it's not trivial to know if the position is a mate or a draw.

Well, you just gave such an example on your own already. Michael's position : draw:

[d]3k4/1p3n2/1Pp3p1/2Pp1p2/3P1P2/4P3/4K1N1/8 w - - 0 1
viewtopic.php?p=822537#p822537

your modification of it : mate:
[d]3k4/1p3n2/1Pp3p1/2Pp1pP1/3P1P2/4P3/4K1N1/8 w - - 0 1
viewtopic.php?p=822592#p822592

Of course, a fortress, engines don't understand at once, seldomly is trivial, e.g. Michael's one isn't at all, still haven't found an engine giving an eval of balance with empty hash, and most of them even aren't able to keep the low eval in hash at Backward neither.

Without engine I can give evalutations of my own as for non trivial fortresses normally only by estimation too, of course. Yet, fortresses are in the double sense of the word just that, bastions of chess, where humans and engines do best in team, each part of the team not as reliable on its own alone as both together.