by Masato Hiramatsu & Mic Sokofun by GAMES 4 BRAINS

In 1998 Masato Hiramatsu (Japan) and Mic (Germany) started a mail-conversation about Sokoban and level-design.
We thought the content of the mails could be interesting for all sokoban-lovers.

So, with friendly permission of Masato Hiramatsu, we like to publish the most interesting parts of this mail-exchange.

Important note:
Mic's contributions are starting with a ">"!

Date:    20.02.98

------- about Gameboy versions and small leveldesign -------

"boxxle" and "boxxle 2" are same as  Japanese Gameboy versions
"sokoban" and "sokoban 2".

Large stages tend to be sparse, tiresome one,
but well-designed small stage is tricky, tight problem and very

Designing such stages is harder than solving it. Mr.Murase's
research, auto-generating sokoban problem is remarkable
approach. His stages generated by computer program (most levels
of twink's web page) are 8x8 and have only 3 packets. But some
of them are very hard.

------- about official screen-size -------

Japanese official(?) screen size is 32x20 and some levels of
"sokoban perfect" and "sokoban revenge" are designed in
full-size. These large levels are omitted for other platform
versions because of resolution of graphic hardware (I suppose).

Some of old IBM-PC version (called "original") have reduced
width, and they are a little different from corresponding levels
of  "sokoban perfect". The reason for this size-reduction may be
raw resolution of old IBM-PC.

Date:    25.02.98

... I appreciate your original work and I have great respect for
your originality. I'm often inspired by excellent concept of
existing levels, too.

------- about improving existing levels -------

Sometimes improving existing levels itself is challenging
motivation and good training for designing skill. My levels in
...  are arranged levels of old sokoban. They
will be harder than original levels and challenging problems for
experienced player.

Sokoban "revenge" means "players' revenge", for all levels in
"sokoban revenge" are created by Japanese sokoban players and
sent to Thinking Rabbit.

You will see many levels of "sokoban perfect" are improved in
"sokoban revenge". Tracing their difference is very interesting.
I recorded similar stages and their differences in comment field
of screen files.
------- about SOKOBLUE (program) -------

I appreciate sokoblue for
recording-solution function, as well as original levels. This
function is essential for designing levels, but few
sokoban-clone have it.
------- about SOKOBLUE and SOKOWAHN (levelsets) -------

I spent about one month for each 100 levels. Roughly speaking,
last 10 levels of my answer files in timestamp order indicates
best 10 of difficulty. They are as follows:


sokoblue2 (sokowahn):


Followings are my play statistics of "sokoblue" and "sokoblue2".
To output such statistics is built in my implementation. Each
number in the table means moves of player recorded in answer
file. These records are not good enough because mouse operation
function was not completed and moving packets by
clicking-packet-and-goal was not optimized until last summer.


100/100 (100%)

total  41011 moves
min       32 moves (level 8)
max     1505 moves (level 57)

      since  Thu May  8 02:30:17 1997 (level 1)
last solved  Sat Jun 21 21:18:39 1997 (level 83)

   |    0    1    2    3    4    5    6    7    8    9
  0| ----   47  134   54  102  116   38   54   32  202
 10|  113   80  134  181   38   85  260  253  264   77
 20|   67  300   81  291  328  100  463  351  472  321
 30|   38  363  116  226  412   78  120  368  284  369
 40|  130  542  125  239  559  293  592  656  526  115
 50|   92  314  410  353  677  333  351 1505  304  530
 60|  131  539  400  435  365  417  397  488  843  767
 70|  259  329  221  311  934  515  427  623  587  991
 80|  127 1078  763  866  750  975  532  768  272 1093
 90|  458  504  964  323  615  505 1442  492  817  961
100|  669


Sokoblue2 (= Sokowahn)

100/100 (100%)

total  49041 moves
min      129 moves (level 2)
max     1337 moves (level 88)

      since  Sat Jun 21 21:33:28 1997 (level 33)
last solved  Wed Jul 30 21:48:06 1997 (level 81)

   |    0    1    2    3    4    5    6    7    8    9
  0| ----  134  129  215  152  153  173  336  323  242
 10|  542  245  195  553  313  668  468  137  216  440
 20|  299  226  341  198  139  320  365  255  455  277
 30|  238  452  863  326  363  489  479  308  248  263
 40|  368  320  271  510  400  326  415  172  361  416
 50|  647  412  534  417  194  870  940  519  758  283
 60|  741  206  196  486  572  487  445  564  554  837
 70|  519  687  434 1153  845  764  637  580  382  823
 80|  802  596  660  439  448  475 1074  663 1337  624
 90|  215  641  630  450  531  406 1112 1061 1240 1061
100|  993

Date:    27.02.98

------- about THINKING RABBIT, level authors and SOKOBAN versions-------

> Was it "Thinking Rabbit" who created all the first sokoban-levels?
> What is his real Name?

I heard Thinking Rabbit is a small softhouse in Takarazuka city,
Japan. Their trademark is designed in perfect #45.

They released some adventure games and other puzzle games
like sokoban. But I think no one exceeds original sokoban
series, for sokoban is best for its simple rule and incredibly
complicated contents.

I tried collecting information of authors from manual booklet of
sokoban perfect and the book "The Sokoban" (1987), in order to
provide proper author field to screen files of sokoban perfect.

The result shows that some levels of sokoban perfect were
created by sokoban players and sent to Thinking Rabbit, but
authors of many levels are still unknown. It is said that most
levels (especially old levels) were created by Hiroyuki
Imabayashi, president of Thinking Rabbit.
They released Windows/Macintosh version of perfect (simply
called "sokoban") (1995) and revenge(1996).

Recently they released "ultimate sokoban" (playstation version,
1996), "sokoban basic" (playstation version, 1997) and "sokoban
selection" (1997, anthology of perfect and revenge).

Some new well-designed levels were found in "ultimate sokoban"
... But new levels in others are too poor
compared with former perfect and revenge. ... You will be
disappointed at them like me. Your work is much better !

------- about creating unique levels and COPYRIGHT -------

I know your principle of creating original levels, of course!
I am tired of collecting freeware/shareware sokobans found by
search-engine, which have same old levels. Many of them claim
their own copyright or shareware fee! Some of them have original
levels, but most are poor designed ones. Your work is
outstanding and I feel your sincere creation through your
levels. I cannot overemphasize this impression.


------- about optimizing solutions -------

Optimizing solution needs different strategy compared with
solving levels simply. I was interested in solving or creating
levels, not optimizing solution like you. But recently I think
some kind of optimizer is necessary for output of current solver
program (it sometimes shows very foolish moves).

Original level data of perfect and revenge had limit of moves.
I omitted this information, but some levels were hard for its
limit in original product.

Date:    03.03.98

------- about favorite games -------

> Easy to learn but hard to control, that's what I like about a good game. And
> talking about great PC-games, here are my favorites besides Sokoban:
> LodeRunner (old version for the C64, I created 32 "new" stages for it),
> Boulderdash, Tetris, Freecell, Zenji (C64-version), Tower Of Babel (AMIGA,
> about 40 stages by Mic, is there a MS-DOS-version?), The Sentinel (great idea,
> AMIGA), Bandit Kings (very atmospheric and entertaining, from Japanese
> softhouse KOEI
My favorites are:

I bought Amiga just to play this game!

Solomon's key
Original version is arcade game by techmo. NES version and
Gameboy version ("Solomon's club") are released. Very nice
puzzle and action.

troddlers (Amiga)
solomon's key + lemmings


I have too many topics to talk about Lode Runner! I will send
them in another mail (wait a moment, please). Concerning sokoban
and Lode Runner, ...

------- about small and compact stages -------

> I don't know, whether to love or hate these stages, because they are not
> solvable by strategy. I know of a sokoban-loving woman, who refuses that kind
> of challenge. On the other hand: Those levels bring variation into gameplay.

The only thing I can point out is: this kind of stages are hard
to design, for designer has no advantages against player. One
solution is testing with solver. Mr.Takahashi's "scrambled egg"
(8-5 of twink's web page, and I included it in "okiyasu" #22) is
incredibly hard small stage. I heard he designed this stage
trying with his solver program.

Date:    09.03.98

------- about level titles and comment-fields -------
I think comments of author (and title of
stage) are attractive factor for player and should be recorded
even few words.

------- about sokoban-level rating system -------

> I have my own rating system, which
> looks as follows: difficulty: 1-10 (1= very easy, 10=near impossible), fun
> factor: ++, +, o, -, -- (++ means much fun/good idea, -- means boring/no fun).

------- about separating fun & difficulty factor -------

Separating fun factor from diffculty is good idea. I have paid
little attention to the former so far, for I thought very
difficult stage essentially has excellent trick and solving it
is challenging. But difficulty is not necessarily generated by
trick itself. Now I am interested in designing artistic initial
and complete pattern, such as line, square, symbol, letter, etc.
Artistic pattern is fun for player. Handy examples are perfect
#176 and #201.

------- about level order in levelsets -------

> I hate Sokoban-versions, which force you to solve one level after another!
There should not be such restriction for selecting levels. My
play style is to select level randomly, and leave it for later
trial if I cannot solve after a few minutes. But this play style
leaves few very hard levels finally. I experienced it in perfect
and revenge. It was really hard...

> By the way: I gave it up, to put levels in order determined by difficulty. Ok,
> the first screen should not be the hardest and the last one should not be the
> easiest. How do you think about this?

I think there are many kind of difficulty and it is impossible
to measure difficulty by single criterion. First level should
enable player to learn basic techniques and final level should
be something great (not in size). Final levels of perfect and
revenge are large and normal solution needs over 10000
moves. Perfect #306 is well-designed and worth for final stage
but revenge #306 requires repeating work only.

------- about misdirection in leveldesign -------

> (++). The dummy is a mean addition! Great!

Sometimes misdirection is important factor. Typical and very
effective example is revenge #XXX (I don't reveal level number
not to spoil your interest !).

------- about small and simple leveldesign -------

> I go for simple screens more and more. You will see it, when you compare old
> SOKOWAHN-levels (You will get them with SOKOFUN!) with what I do now. Old
> levels might have been harder sometimes, but not better in my opinion.

1st prize of last sokoban contest was as follows (by Hiroshi
Yamamoto, I included this in "ps" as #100). Comment of reviewer
is "simple is best".

--------------------------- cut here ---------------------------
#   #
line omitted to avoid copyright infringements
#.     #
#  * * #
# * # ##
### @ #
--------------------------- cut here ---------------------------

Stages with only one packet (revenge #80 - #84) provides problem
of finding route and there are no other difficulties. Revenge
#14 proved that stage with two packets can generate some other
problem (this is the only stage with two packets and designed by
Hiroshi Yamamoto, same author as above stage). Mr.Murase's
auto-generated levels (and his handmade levels, revenge #7 and
#8) have three packets. Recently he is trying joining small
stages generated by program together into new level. I also
think they are available as basic puzzle unit.

I often find simplifying existing level into core trick result
in very easy problem. My current motivation is concatenating
basic puzzle pieces systematically and generate new meta-puzzle.

------- about keeping tention, mean traps and hard sokoban-levels-------

My another motivation is keeping tension in middle or large size
stage. Joining small hard level sometimes generates another
unexpected easy solution, which is common trap for level
designer. I am trying avoiding it but no effective solution is
found so far.

I think #275 and #276 of perfect are
bottleneck for many players. But ultimate trap in original
"sokoban perfect" is as follows:

Challenging message from Thinking Rabbit for sokoban perfect

"There is only one stage that can never be solved in 306 stages.
What is it ?"

Many players were entrapped. But the answer was all stages can
be solved ...

Date:    17.03.98

------- about leveldesign for the "common player" -------

> I found many of the small levels interesting. Beginners might be trapped
> sometimes. The problem, which I am thinking of is: How to create sokoban-
> stages, that are interesting for the common-player. The common player is not
> very patient. He will not love too large screens. The common player is
> interested in initial problems more than in final difficulty. He dislikes
> frustration (Who likes it?). You will not believe it, but over 80% of incoming
> sokoblue-orders include movies!
> It is also a problem for many players (me included) when goals are dispersed
> all over the screen (as done in some of my symmetricals). Sometimes things get
> so intricate, that fun-level is decreasing.

Sokoban is very popular in Japan (and in other countries), but
true number of players seems very few. Some stages of perfect
and revenge are too hard and I think many people gave up
completing them. Sokoban released recently by Powerhouse has
very easy stages, but player can learn nothing from them.
Moderate difficulty is desirable but to achieve it is

------- about essential rules for level designers -------

My goal is not designing large stage, but introducing new
paradigm. Though there are many variations, classic style stages
are summarized as two techniques:

* re-arrangement of packets
* filling goal

Interaction of only two goals brings new complexity shown as
sokoplay #19 arrange ver.2 above. I tried interaction of three
or four goals in my recent stages, but they seem too hard (hir
#16 and hir #17, 9-4 and 10-5 of twink's web site).

Date:    26.03.98

------- about trimming spaces -------

> Sometimes I leave space in my screens (so in sokoplay #98), even when there is
> no need for it. Too much space can add to gameplay (players can be fooled!) or
> look  better.

I realize such kind of looseness (and I often leave spaces). My
intension for eumerating such point is not recommending you to
design so but calling your attention. Usually my first step of
tuning process is trimming space or adding more packets and
goals. "Trivial tune" means I tried to make stage harder but my
efforts were without result and left trivial changes only. You
can start tuning by yourself from these trimmed pattern.


------- about Masato Hiramatsu's activities as sokoban level designer-------

> Have you contributed to "perfect" and "revenge" or to another commercial
> product?

I started creating sokoban levels in August 1996. I sent two
levels (hir #1 and #3) for last sokoban contest (1996 autumn,
after playstation version "ultimate sokoban") but sokoban
products released from PowerHouse after contest contain old
levels only. "Ultimate sokoban windows version" was released
about two weeks ago. I will examine levels but they seem to be
same as playstation version.

I want to share my levels with other sokoban players, on the
other hand I will enter my levels for next contest, which means
all rights are sent to PowerHouse. That is ambivalent.


------- about SOKOFUN -------

I downloaded "SOKOFUN". Nice work! This version should be
praised for many features and original levels.

I realized exact rule of your two games (Sokoblue and Sokowahn) by English help. I learned German as second foreign language at university, but I
can hardly comprehend it...

------- about personal sokoban style -------

> I avoided screens with more than one goal(area), to make solution-planning not too
> confusing. Exception sokoplay #12, where principle was broken, but structure
> of problem is not complex. Leveldesign with more goals needs space. 15x15
> matrix as given in sokoblue (my wish!) is not large enough.

Obviously our starting points are different. When I started
creating sokoban stages, I was challenging hard stages of
perfect and revenge and my interest was how to introduce new
design concept, which will bring new kind of difficulty. Large
screen size did not matter for easy mouse operation (small size
is better, of course). I feel sokoplay #12 has very unique

Date:    16.04.98

------- about level-pirates and others -------
After sokoban revenge, many levels must be sent to Thinking
Rabbit. I hope my levels will be included as a part of new
official sokoban release some day.

I saw worse situation outside of Japan. There are too many
clones of classic levels, which indicates programming simple
sokoban game system is very easy but designing levels is not so
easy. Other levels of gameboy versions are collected in
Macintosh version, but they should be a part of perfect/revenge
and lack of title, author and comments. There are few original
levels but most of them are not worth playing (except for

I examined Pocoman levelpack "Green Machine" a few days ago and
found most levels are copied from existing levels and simply
concatenated without any original concepts. It is nothing but
stealing and there is no creative work.

------- about creating new sokoban-levels -------

> I never tried to tune a certain (original) level. But sometimes I had a
> certain level or part of it in mind, when I created one of my own.
> Of course I'm interested in making hard stages, but it is fun as well to find
> out new ideas. This brings in variation.
> Many people should have fun by solving my creations. So variation in design
> and difficulty is essential.

Recently most of my designing process is tuning work. Examining
prototype level again and again shows the key point of the level
and sometimes indicates unexpected new concept.

But this style needs effective turn-around of editing/playing
and requires many hours. I developed my version of sokoban for
this purpose and I am still planning more features. After
examinig sokoplay levels, I will return to programming work. I
have many levels of my own to be tuned but I don't have enough

------- about original game concepts -------

> Original game concepts are hard to find nowadays. 95% of commercial production
> seems to be boring for me. Just like you I have to set priorities, cause my
> time is limited. So I concentrate on very few games.

Recently, gamesofts have high-quality graphics, sound and story
like novels. But what I want is basic excellent concept. I heard
LodeRunner II is coming soon, but I cannot expect enthusiastic
hours of old happy days.

Date:    18.05.98

------- about the need of competent testers -------

>  Sometimes other players solve my
>  levels much more efficiently than me. I suspect unexpected easy
>  solution if an other solution is too short.
>  Unexpected short solutions can be tricky too. Movie-files of other players can
> be helpful too avoid easy solutions. My problem is: I know very few persons,
> who can help me doing test-work.  

That is why I am trying to make detailed report. It is hard to
find useful commentator for such kind of problem.

------- about SOKOPLAY levelset -------

100/100 (100%)

total  32471 moves
min       35 moves (level 9)
max     1509 moves (level 94)

      since  Wed Feb 25 18:55:44 1998 (level 1)
last solved  Sat May  9 15:31:59 1998 (level 100)

   |    0    1    2    3    4    5    6    7    8    9
  0| ----  196  250   79   78  102  324   96   80   35
 10|   86   71  103   75   72  116   85   67   80  196
 20|   51  113  124  719  146  118  264  196  206  909
 30|   72  189  128  234  240   43  144   58   64   54
 40|   48  413   96  105  112   83  163   92  392  130
 50|  100  130  706  511  327  130  294   68  134  320
 60|   58 1333  652  748  118  506  282  900  723  524
 70|   89  370  139  265  137  358   91  350  947 1096
 80|  645  839  759  104  878   93  495 1241  222  181
 90|  152  103  299  636 1509 1304  847  531  870  326
100|  134

Date: 03.06.98

------- about "SOKOBAN perfect" masterpieces and solving levels on paper-------
> I feel great respect for all contributors. They have given me so much
> fun, especially when I think back to the old days, when I had my first
> contact with the game solving those famous "original" 50!

Me too. I wondered at many excellent ideas and they gave me
motivation for new stages. I want to play their new works but
few outstanding stages are found after sokoban revenge.

> There was that incredible last stage #306. You were right: This one is
> really well designed, and I should take back my judgement about large
> screens. #306 is an very clever arranged collection of swingdoors. The
> keypoint is to make possibbe redirection in the lower right. This screen
> was the only one, that I had to print out and after calming down a bit,
> solve on paper.

It is the last stage for me, too. I felt this one is too large
and left it for later challenge. I sometimes challenged this
stage while completing other stages, but finally this stage is
left. I spent many hours in order to push the first packet into

> The second to last was #300 for me. This one had a real sophisticated
> stocking-problem. I was stuck on it for some time, but then I made a
> plan on paper. This led to success immediately.

I think the difficulty of this stage is narrow entrance and
well-designed stock pattern. Initial pattern of packets in left
region was also hard for me.

> Another masterpiece was #302. It took me more than 60 minutes to solve,
> just as #303.

#302 is the final stage of many early versions (sokoban 3, NES
version, MSX version, GENESIS version, and so on). This stage
requires walking around from right to left many times, but moves
limit is severe (3500). I had to optimize my solution a little.

#303 (by Seiji Nishikawa, programmer of MSX version) first
appeared in MSX version ("sokoban toolkit" 1984). This stage and
#288 are perhaps oldest "unfinished stage" except for the trap
level #84. Hiroyuki Imabayashi said perfect #303 was the only
stage he couldn't solve and therefore this stage was not
included in Thinking Rabbit version until sokoban perfect.

The End

BACK to GAMES 4 BRAINS  homepage