# This is a collection of blunders of GNU Go in Stefan Mertin's private
# 13x13 - Computer Go tournament in 2001 (see
# http://www.geocities.com/comp_go/13x13.html). I think they should be
# quite useful as tests. 13x13-positions are specific as problems, and
# they are quickly evaluated both by GNU Go and by us when maintaining.
# (And of course, these are games against some opponents we would like
# I also added a few test cases where GNU Go played right in the game,
# but went wrong in a replay by a newer version (3.14 / 3.15).
# Most of the variations in the sgf-files are due to Stefan Mertin.
#DESCRIPTION=M2 has the huge follow up of rescueing H3, not seen by GNU Go.
loadsgf games/mertin13x13/goliath-gnugo1.B+11.sgf 22
# E10 is a little better but the difference might not be worth regressing
loadsgf games/mertin13x13/goliath-gnugo1.B+11.sgf 36
loadsgf games/mertin13x13/goliath-gnugo1.B+11.sgf 58
3 restricted_genmove white E2 E5
# Of course it is nice if you can get away with J11 as GNU Go did in the
# game, but I think L9 is urgent.
loadsgf games/mertin13x13/goliath-gnugo2.W+9.sgf 30
loadsgf games/mertin13x13/goliath-gnugo3.W+0.sgf 42
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/goliath-gnugo3.W+0.sgf 44
loadsgf games/mertin13x13/goliath-gnugo3.W+0.sgf 48
loadsgf games/mertin13x13/goliath-gnugo3.W+0.sgf 54
8 restricted_genmove white K6 K7 J8 H8
loadsgf games/mertin13x13/goliath-gnugo3.W+0.sgf 72
loadsgf games/mertin13x13/gointellect-gnugo1.B+14.sgf 12
#CATEGORY=TERRITORIAL_EVALUATION
loadsgf games/mertin13x13/gointellect-gnugo1.B+14.sgf 20
#CATEGORY=TERRITORIAL_EVALUATION
loadsgf games/mertin13x13/gointellect-gnugo1.B+14.sgf 24
#loadsgf games/mertin13x13/gointellect-gnugo2.W+8.sgf 22
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gointellect-gnugo2.W+8.sgf 28
# New failure (comparing 3.1.15 and 3.0.0)
loadsgf games/mertin13x13/gointellect-gnugo2.W+8.sgf 32
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 24
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 32
# This got broke between 3.0.0 and 3.1.14.
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 42
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 48
19 restricted_genmove white M8 N3
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 60
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 66
loadsgf games/mertin13x13/gointellect-gnugo3.B+1.sgf 94
loadsgf games/mertin13x13/katsunari-gnugo3.B+14.sgf 16
# At least try to connect.
loadsgf games/mertin13x13/katsunari-gnugo3.B+14.sgf 34
loadsgf games/mertin13x13/katsunari-gnugo3.B+14.sgf 64
loadsgf games/mertin13x13/katsunari-gnugo3.B+14.sgf 66
#CATEGORY=FUSEKI_STRATEGY
# Too early for 3-3 invasion at L11 with both UL and LR corner open
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 10
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 12
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 20
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 32
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 64
# A nice tesuji by GNU Go worth remembering!
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 74
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 82
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 102
loadsgf games/mertin13x13/katsunari-gnugo1.W+45.sgf 110
loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 12
loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 20
loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 44
# The followup of K13 is undervalued.
# In fact, there is none at all (3.3.8)
loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 50
# C8 can be cut off. Stones added to make the testcase more clear cut.
loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 60
41 restricted_genmove white C8 D7
loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 126
#? [N3|N2|L4|L3|L2|L1|M1|N1|M4]
# A correction in this shape would need subtle tuning, as leaving
# the gap at D7 (by playing D6) should often be bad.
loadsgf games/mertin13x13/gnugo-goliath1.W+11.sgf 25
# To get better evaluation for B6 one would need to
# 1) add the follow-up value of the monkey jump (very high here as
# 2) move the influence barrier in the current situation around B6
# a little further down, and/or add a reverse followup value here
# Part of the overvaluation of M11 is due to the huge shape factor (+8.00)
# by pattern CC77, which does not make too much sense here.
loadsgf games/mertin13x13/gnugo-goliath1.W+11.sgf 27
# B4 is really huge here, appr. 20 pts.
loadsgf games/mertin13x13/gnugo-goliath1.W+11.sgf 31
# L3 is 2pt double sente.
loadsgf games/mertin13x13/gnugo-goliath1.W+11.sgf 85
#CATEGORY=FUSEKI_STRATEGY
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 7
47 restricted_genmove black K8 K9 G11
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 11
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 13
# Renumbered since there were 2 test 49's
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 19
# Here N11 is generated as an "additional attack/defense move", although
# the black dragon is still dead after black connects at N11.
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 27
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 33
# GNU Go 3.1.15 wants to play F10 here (although 3.0.0 got H10)
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 35
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 61
# This is very difficult. There are some variations in the .sgf-file,
# most of them due to Stefan Mertin. Of course J13 (played by GNU Go)
# is an annoying blunder.
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 67
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 83
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 99
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-goliath2.W+38.sgf 111
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-goliath3.B+16.sgf 55
# Moves like C8 are severly undervalued by GNU Go. /ab
loadsgf games/mertin13x13/gnugo-katsunari1.B+21.sgf 13
59 restricted_genmove black J11 C8
loadsgf games/mertin13x13/gnugo-katsunari1.B+21.sgf 15
loadsgf games/mertin13x13/gnugo-katsunari1.B+21.sgf 75
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 13
# New failure (comparing 3.0.0 and 3.1.15)
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 25
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 33
# New failure (comparing 3.0.0 and 3.1.15)
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 35
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 47
#CATEGORY=ATARI_ATARI + ENDGAME_TUNING
# New failure (comparing 3.0.0 and 3.1.15)
# EB1404b produces nonsense here.
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 77
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 79
#CATEGORY=TERRITORIAL_EVALUATION
# There is s.th. odd in the influence function in comparison between
loadsgf games/mertin13x13/gnugo-katsunari2.B+66.sgf 89
69 restricted_genmove black D4 D5
#CATEGORY=FUSEKI_STRATEGY
# I think that securing the corner is larger.
loadsgf games/mertin13x13/gnugo-gointellect1.W+28.sgf 11
loadsgf games/mertin13x13/gnugo-gointellect1.W+28.sgf 19
loadsgf games/mertin13x13/gnugo-gointellect1.W+28.sgf 29
# New failure (comparing 3.1.15 with 3.0.0)
# The "attack last move played, although it seems bad" heuristic sometimes
# produces very bad choices.
loadsgf games/mertin13x13/gnugo-gointellect1.W+28.sgf 31
#CATEGORY=OWL_TUNING, SEMEAI
# The "attack last move played, although it seems bad" heuristic sometimes
# produces very bad choices. Also, owl is wrong here.
loadsgf games/mertin13x13/gnugo-gointellect1.W+28.sgf 33
#CATEGORY=TERRITORIAL_EVALUATION
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 13
75 restricted_genmove black D11 E12
#CATEGORY=TERRITORIAL_EVALUATION
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 17
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 19
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 21
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 23
# M2 is better for eye shape here.
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 41
loadsgf games/mertin13x13/gnugo-gointellect2.W+2.sgf 55
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 11
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 23
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 25
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 29
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 45
#CATEGORY=TACTICAL_READING
# GNU Go claims that D6 defends the worms D9 and F9
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 65
# New failure (comparing 3.1.15 with 3.0.0)
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 69
88 restricted_genmove black A9 D6
# New failure (comparing 3.1.15 with 3.0.0)
# (I hope I read this out correctly. See variations in sgf-file.)
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 83
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 99
loadsgf games/mertin13x13/gnugo-gointellect3.W+37.sgf 109