Welcome to my website on retroanalysis!
This website is under construction, and currently it hosts only my most recent research project on "mushikui reconstruction".
The word "mushikui" came from the mushikuizan puzzle, and it's a Japanese term meaning "eaten by bugs".
Mushkikui reconstruction is a type of chess game resconstruction task, proposed by me on the retro mailling list in 2011. In each problem, solvers are given a game score, written in PGNep format (explained below), but all the characters are concealed by "*" symbols, so that only the length of the representation of each move is known to the solvers. The task for the solvers is then to recover the entire game score based solely on these informations. All the problems need to have a unique solution in order to be considered as valid.
The format I'm using in mushikui reconstruction is what I call the "PGNep" format, which stands for "PGN spec" + "ep". This means that everything in this format is the same as described in the PGN spec document, with only one exception being that I use "ep" to denote en passant move. The main reason why I choose my format this way is because most people in this field seems to get used to put "ep" whenever they write a game score, so I believe this format will make things more "natural" to the people in this field.
Summary of this format:
En passant moves are denoted by "ep". For example, "cxd6ep".
Check moves are marked by "+", and checkmates are marked by "#".
No special notations are used for double checks, discover checks, stalemate moves, etc.
Use "=" to denote promotion. For example, "axb1=Q+".
Disambiguation is needed. For exmaple, use "Nb1xc3" when there're knights of the same color positioned at d1 and b5, and all the three knights can make capture at c3 legally at that momemnt (that is, they're not pinned).
One subtle situation is when one of the moves will give a discover check while the other will not. In this case although we humans can distinguish the two moves by simply noticing if there's a "+" mark right there, but according to the PGN spec 18.104.22.168, we will still use disambiguation notation in this case.
All the problems on this webpage have a unique solution, and are computer checked. I don't consider any sequence that doesn't have a unique solution, even if it does force something, for example its final position or certain moves, etc.
Some of my compositions
(2011.5.17.1) This one was my very first mushikui composition, but it's pretty tough to solve.
First example with Valladão theme, and also the first example that forces promotion to rook.
Notice that if we add 12...**** to the problem, then we obtain a sequence with a Schnoebelen promotion to rook. The solution is 12...Kxc8.
(2011.5.23) "Cradle to Grave". A long, challenging problem, but still with reasonable difficulty for human solvers.
This one achieves the following: (1) Valladão theme on both sides (2) both O-O and O-O-O castling (3) forces promotions to queen and bishop (4) forces the game to conclude (draw).
Specially thanks to Mario Richter for providing the segment 1.** *** 2.** *** 3.** **** 4.***** **. Without him this cool problem cannot be done.
The portion up to the 35th single move is a stage, which is also the shortest sequence I know so far that forces the O-O move (actually it forces both castling alreday).
First one that forces a one-step pawn move in the opening, in particular 1.d3. THE shortest such sequence is later found by computer.
This one is also the current single-stage length record holder. It has 39 single moves and 145 stars.
Previous record holder for the shortest sequence that has a disambiguation move.
(2011.5.27) Definitely impossible for human solvers. Even for computers, the running time is almost like forever. This is the first time I use parallel computing just for handling a single problem, and still it took a whole day to finish checking the validity of this one, using the older version of my program. With the new version, it only takes less than 40 seconds to check this one, but still this one remains the toughest problem for computer programs so far.
Achieves the following:
(1) New record for the longest sequence.
(2) Up to move 42.5 is a single stage, setting the new record for the longest single stage.
(2) 22 consecutive non-capturing moves, starting from 32.
The part from move 32. was found by genetic algorithm.
Using the old version of the checking program I wrote, I examine all the possible mushikui sequences up to six single moves. Among them, only two will force a unique solution:
(2011.5.19, M.-T. Tsai + Computer) All the valid sequences up to 6 single moves:
Later, Mario Richter (2011.5.26) first determined all the valid sequences with 7 or 8 single moves that doesn't contain a disambiguation move. After I rewrite my program, I then completely determined all the remaining valid sequences. The results are shown below. Each of them is a pretty good puzzle to human!
(2011.6.4, M. Richter + M.-T. Tsai + Computer) All the valid sequences with 7 single moves:
The last one is THE shortest sequence that has a disambiguous non-capture.
All valid sequences with 9 or 10 single moves is later completely determined as well. See the data page for details.
In my opinion, I think a problem of this type is more enjoyable when it consists of a lot of stages, with each stage having a unique solution on its own. For example, Problem (2011.5.19) above can actually be decomposed as follows:
In this decomposition, the end of each line represents a stage that has a unique solution. Totally it has 20 stages and 56 single moves, so I shall define its "stage density" to be 20/56. This is the mushikui sequence I know so far that has the highest stage density.
Sometimes, it is possible to replace some of the moves by question marks "?", to indicated that the move is completely unknown, not even the length of the representation of that move, but still forces a unique solution. For example, the shortest one that allows a question mark is 1.*** ? 2.**** ****, whose solution is still 1.Nc3 d5 2.Nxd5 Qxd5. Another example is 1.? ? 2.**** **** 3.? ***** 4.**, whose solution is 1.d4 c5 2.dxc5 Qa5+ 3.b4 Qxb4+ 4.c3. This one has 3 question marks and 7 single moves, so I shall define its "question mark density" to be 3/7.
Some open quests
Here I shall propose several open quests for everybody. In the following, by "sequence" I mean a mushikui sequences with a unique solution. The "length" of a sequence could mean one of the following: number of single moves, or number of stars "*".
Find the shortest sequences with 1.a3, 1.b3, 1.e3, 1.f3, 1.g3, and 1.h3.
Determine all the sequences with eleven single moves, twelve single moves, etc.
Find a sequence with the highest stage density. Current record: 16/39.
Fina a longest sequence. Current record: 86 single moves, 302 stars.
Fina a longest sequence that has only one stage (that is, it has the lowest stage density). Current record: 85 single moves, 298 stars.
Find shorter sequences that achieves the same theme as in some of my compositions (checkmate, O-O move, Valladão, promotion to bishop, etc.)
Find a sequence with a consecutive ** ******, but it is not an en passant capture.
Find sequences with other retro themes like Ceriani-Frolkin, Pronkin, etc.
Find a sequence with a Schnoebelen promotion to bishop (queen is not possible, and both knight and rook are done).
Find a sequence with consecutive ** *** **** ***** ****** ******* (2 through 7).
Find a sequence that contains the most consecutive non-capturing single moves (check is OK). Current record: 22 single moves.
Prove that there's no sequence without a capturing move.
Some open quests regarding question marks:
Find a sequence with the highest question mark density. Current record: 7/11.
Fina a sequence with the most number of consecutive question marks. Current record: 4.