Little Golem#1 · shalev
29 Jan 2014
I'm posting a review of this recent game against mohex:
http://www.littlegolem.net/jsp/game/game.jsp?gid=1607975
1. c2
Looking at mohex's previous games, I noticed that it seems to play opening moves that are a bit too strong, and it also seems to not swap enough. This made me think it probably won't swap b2 or c2, in which case I prefer having the c2 stone (it's strictly stronger, although in practice only by a bit).
2. f8
I was right - mohex didn't swap. This isn't necessarily a mistake: Daniel Sepczuk also didn't swap c2 against me, and in fact he also played 2. f8, which is a solid move.
3. j9
This is a standard corner move. I was curious to see how mohex responds to this, because it doesn't seem to follow the standard corner patterns.
4. h7
I thought this was a pretty bad move, mostly because 5. j6 is such a strong response to it.
5. j6
The idea here is that j6 and j9 together cut up the right side into segments that make it hard for mohex to connect efficiently. j6 is also on the "third diagonal", which means I'm guaranteed to get a ladder on the top (and this indeed happens). j9 is connected to the bottom, and j6 almost is (I didn't calculate exactly).
6. j5
This move tries to block j6 from connecting to the top, and in the process gives me a 6-th row ladder. It's a decent but predictable move.
7. e7
This was my plan against j5: the e7 stone basically acts as a ladder escape fork. Note that it "diagonally reaches the top":
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7f5d6e4c5d3b4
while threatening to connect to the bottom via d9. As I explained in my last annotation, the c2 stone makes e7 feel a lot like a 5th-row stone. It almost feels like I have this:
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7c1b2b1a2a1d2d1e2e1f2f1g2g1h2h1i2i1j2j1k2k1l2l1m1
If I had that, then e7 would indeed escape the (now 4th-row) ladder.
8. f6
This is a decent move that makes use of the fact that e7 is not actually a 5-th row stone. My first instinct is to respond like this:
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7f6d6e5c5
However, this fails to escape the ladder:
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7f6d6e5c5e6d7d9i6i4i5i3g6h5h6f7g5g4
Similarly, d7 also fails:
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7f6d7f5e5d9i6i4i5i3g6h5h6f7g5g4
9. i6
Instead of playing next to the e7 stone, I decided to force mohex to decide if it's going to play i4 or i5 once I start my ladder. My escape strategy at the top left corner will depend on this choice.
10. j7
This move surprised me a bit. I didn't get it's purpose at first - it seems to try to dispute the connection of j6 to the bottom, but at first I didn't see the reason to play this now rather than later.
11. i7
I think this was almost forced, because I want j6 to try to connect to the bottom.
12. i4
After seeing the move, I felt like I understood mohex's 10. j7 move better. It j7 provides some strength for ladders on the right, like this:
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7f6i6j7i7i4i5i3k3j4l4k5
Mohex played j7 first, to see if I respond with i7 or k6, so that it could play i4 in the former case and i5 in the latter. This is a clever trick, but it doesn't do much, because I wasn't really planning to give mohex a ladder on the right anyway.
13. h5
14. h4
15. g5
16. g4
Over the course of these moves, I calculated a strategy: ladder up to g5, then play e5.
17. e5
I think at this point I've almost certainly won. The e5 piece is connected to the top. If mohex blocks at f5, I win with d9, as happened in the game. If mohex instead plays something like d9, then I connect at f5, and although this chain is not quite connected to the bottom, it very nearly is. I don't think it's difficult to complete this connection. For example:
http://www.trmph.com/hex/board#13,c2f8j9h7j6j5e7f6i6j7i7i4h5h4g5g4e5d9c8d10f5h6i5h9i8i11k10k11j11i13i10h11h10g11g10f11i12h13h12g13g12f13f12e13d12e12e10f10f9g9g8h8g7
18. f5
19. d9
This move may look surprising, but it wins, as the game shows. It's probably one of the only winning moves.
20. e6
21. c7
The c7 piece is now connected to both the top and bottom, making it the first piece which is guaranteed to be part of the winning path.
From here the game goes on for 50 more moves, because mohex doesn't resign. If you want to see why c7 is connected to the top and bottom, check out the actual game.
Overall, the interesting part of the game was shorter than in my games against good human players. It's hard for me to estimate mohex's rating from this alone, but I don't think it can be much over 2000. I'd guess that it will end up being between 1900 and 2100 somewhere.
It's also hard for me to pinpoint to exactly what mohex's mistakes were. 4.h7 was probably the biggest mistake, but mohex also played a bunch of boring, predictable moves when it should have tried crazy things to get out of a losing position. Also, not playing a corner pattern at the bottom right was surely a mistake as well.
I'd be happy to answer any questions you have. Let me know what you think!