The New Sudoku Players' Forum

[champagne]

I think that this topic is worth a separate thread with at the end a code doing the job.

I and others made in the past vicinity search in several cases

. The pattern game where the given cells could not be changed (many players)
. The search of potential hardest with at that time 18-24 clues given and a limited -1+1 depth.
. The search of new 17 clues (a depth +-3 made by mladen Dobrichev)
...

And several ways to do it have been applied.

I recently started a search of “non degenerated tridagon puzzles” with high ratings
http://forum.enjoysudoku.com/non-degenerated-tridagon-puzzles-direct-search-t45331.html

The reference is the ” loki family “ derived as far as we know from the “loki” seed with an impressive number of high ratings.
If we just consider mith’s file of T&E(3) puzzles, this is

4 634 322 puzzles
67 539 solution grids
Around 68 puzzles per solution grid.
Most of them produced using several “vicinity” processes.


I have an ongoing scan of the first 33million grids of the catalog as first target.
I expect to get more than 400k puzzles rating SKFR>=10.5

This is as 400K “loki” starts to use as seeds to find other “loki” families.

The first 20 million puzzles produced

332657 puzzles “moderately expanded” in
183135 solution grids.

The goal is now to see if one of these seeds can produce new interesting families as the “loki” one.


This cannot be done seriously without a very efficient “vicinity” search. I have a first version of code doing it, but I open the thread waiting first for external comments.

[champagne]

The first question is” what can we do starting from a seed”. I keep apart the “max expand => new minimal”.

A vicinity search in a general definition can be written

Having a puzzle ‘A’ with a solution grid sA
Find a puzzle ‘B’ with a solution grid sB
Keeping untouched a given property.

By experience, in many cases, small variations around a puzzle having the property are quicker than a direct search of the property.

Here the property is a high SER rating, and, as the seed has this property due to the tridagon threat, the ‘B’ puzzles having a high rating has good chances to have the same threat, although it could be another reason.

If sA=sB (same solution grids) then, both ‘A’ and ‘B’ hit the set of ED unavoidable sets of the solution grid.

If sA!=sB, sA and sB differ by an unavoidable set, and this UA belongs to the set of UAs of sA.
Note : This UA can have subsets, then it would not be in the minimal list of UAs.

Also, if one puzzle is minimum, for each given, you have one UA (or more) hit by this given and only this given.

We see here that the list of UAs of the solution grid attached to the puzzle to process can play a key role in several “vicinity tasks”.



The vicinity can be applied to a non-minimal puzzle, but, here, I assume that only minimal puzzles are studied and that the second puzzle is “preferably” minimal as well.

If ‘A’ is minimal ‘B’ is minimal and sA=sB, then some clues of ‘A’ are replaced by some clues of ‘B’.
The classical vicinity is for -1+1 -2+2 ...

If sA!=sB, then, we still have the possibility to replace clues of ‘A’ by new clues, but we can also keep all the clue cells of ‘A’, changing the value for part of them, adding, if necessary, clues to get a valid puzzle.
One special case of this option is known as “twin puzzle”. In a twin puzzle, all cells of the UA common to sA,sB are in the clues of ‘A’, The clues values of these cells are given by sB


Specifications of the program..
________________________
The vicinity to build must in priority process the output of the “non degenerated tridagon” finder.
This is an expanded puzzle with a high rating. Would be accepted seeds made of a valid puzzle suspected to have the potential to produce searched output.
As the vicinity is usually made in several cycles, the goal is to make all quick moves, with a priority given to small moves.
________________________
The program will find minimal puzzles of the entry, each of them processed separately.
The three groups of ‘B’ puzzles will be produced in the limit of what can be quickly done
‘B’ puzzles will be valid and pass low ratings. The “minimal” property not necessarily checked. In my code, I added this constraint (expensive) thinking that a non-minimal could have redundant clues killing the “property” and that any puzzle in output should later be checked for “minimal”.
______________________________
First group of ‘B’ no cell killed:

The sB solution grids will be built from ED smallest unavoidable sets. All UAs have a minimum of one cell hitting ‘A’, so the number of cells values to change can be one and up to the size of the UA (twin puzzle)
Done on “minimal” puzzles the corresponding ‘B’ are quite like other groups. All ‘valid’ will be kept, adding cells if needed using UAs coming from the new solution grid. The main filter to use is the size limit for the UAs. A max number of seeds to produce can also be considered. And the filter for easy puzzles should be applied.
Note : as far as I know, the twin puzzle has been defined for puzzles in a “max expand” status. Working on “minimal” puzzles, such situations (all cells of the UA in the givens) should be uncommon.
___________________________________

Second group ‘B’ in the same solution grid.

The wider scope is (-n+p), but n>1 is not easy to implement. The program will start with -1+p.
The basis is -1+1 if this is possible (all UAs not hit by the remaining clues have a common cell).
-1+2 can be seen for 3 reasons:
. -1+1 seemed possible with a common cell, but the set of UAs did not cover the field,
. The set of UAs already required a minimum of 2 clues.
. One could also want to use non-common cells to have more puzzles in output
More than -1+2 seems too much for a start.

All these puzzles are discarded If the puzzle is “easy to solve”
_____________________________
Third group, ‘B’ is in another solution grid.

The start is the same: kill one given, get the UAs not hit by the rest.
All these UAs will have the given as common cell, using it would be redundant with the first group.
We must choose one UA to build the corresponding sB. Then, any cell of the UA can be used as +cell and a second one added if necessary if ‘B’ is not valid. More cells are possible, but 2 cells as limit seems reasonable.
This should be done with the smallest UAs and only puzzle passing the low rating filter kept.

[Hajime]

Is your vicinity search of -1/+1
A. kick 1 cell out and replace that cell with another given? ( I used this in the Patterns Game) or
B. kick 1 cell out and any other empty cell can be the cell with some given?

[coloin]

Having done extensive vicinity searches ...I think the best option to extend the vicinity search is to somehow ...
a. opt for new solution grids .. keeping in the same grid can find more puzzles from the same max-expand, but new grids always means new puzzles
b. only exchange 6 clue values ... in these puzzles one of the boxes contains the 011 or 111 triplet of clues these clues don't feature elsewhere.

[champagne]

Is your vicinity search of -1/+1
A. kick 1 cell out and replace that cell with another given? ( I used this in the Patterns Game) or
B. kick 1 cell out and any other empty cell can be the cell with some given?

Let me answer first to A.

What you write is what I did in blind mode when I played in the game.
This is part of what I called the first group, with "no cell killed".
But what is proposed (and done in my draft) is much more

Instead of
"change one cell see if valid", I write
"change one unavoidable set, keep existing clues cells, see if more cells are needed to get a valid puzzle, then check for a minimal puzzle"

If the unavoidable set has only one hit in the puzzle and if the new puzzle is valid then we are in your A.

The so called "twin puzzle" is on the other limit of this A. case :
change "n" cells, see if the new puzzle is valid.

[champagne]

Having done extensive vicinity searches ...I think the best option to extend the vicinity search is to somehow ...
a. opt for new solution grids .. keeping in the same grid can find more puzzles from the same max-expand, but new grids always means new puzzles
b. only exchange 6 clue values ... in these puzzles one of the boxes contains the 011 or 111 triplet of clues these clues don't feature elsewhere.

I agree that more a. should be searched, and this is done here using UAs to produce the first group of puzzles

For your point b.

i first worked on a similar idea but
in a general situation, I don't see which cells should be ignored

This post of blue

Hidden Text: Show

Here's one where the "odd box" in the magic square changes from box 5 to box 2:

Code: Select all
tridagon digits: 1,2,3
magic square: B1245
type 2
+-------+-------+-------+ +-------+-------+-------+
| 4 5 . | . . . | . . . | | 4 5 1 | 3 7 8 | 2 6 9 |
| 6 . 7 | 9 . 5 | . . . | | 6 2 7 | 9 1 5 | 8 4 3 |
| . 8 9 | . . . | . . . | | 3 8 9 | 4 6 2 | 7 5 1 |
+-------+-------+-------+ +-------+-------+-------+
| . 6 4 | . 8 7 | . . 5 | | 1 6 4 | 2 8 7 | 9 3 5 |
| . . . | 6 . . | 4 . . | | 5 3 8 | 6 9 1 | 4 2 7 |
| 9 7 . | 5 . . | . . . | | 9 7 2 | 5 4 3 | 6 1 8 |
+-------+-------+-------+ +-------+-------+-------+
| . . . | 8 . 6 | 1 7 4 | | 2 9 5 | 8 3 6 | 1 7 4 |
| . . . | . . 4 | 3 . 2 | | 8 1 6 | 7 5 4 | 3 9 2 |
| 7 . . | . . . | . 8 . | | 7 4 3 | 1 2 9 | 5 8 6 |
+-------+-------+-------+ +-------+-------+-------+

{-7r7c8,+9r5c6} -->
type 1
+-------+-------+-------+ +-------+-------+-------+
| 4 5 . | . . . | . . . | | 4 5 3 | 7 2 8 | 6 1 9 |
| 6 . 7 | 9 . 5 | . . . | | 6 1 7 | 9 3 5 | 2 4 8 |
| . 8 9 | . . . | . . . | | 2 8 9 | 4 6 1 | 7 5 3 |
+-------+-------+-------+ +-------+-------+-------+
| . 6 4 | . 8 7 | . . 5 | | 1 6 4 | 2 8 7 | 9 3 5 |
| . . . | 6 . 9 | 4 . . | | 5 3 8 | 6 1 9 | 4 2 7 |
| 9 7 . | 5 . . | . . . | | 9 7 2 | 5 4 3 | 8 6 1 |
+-------+-------+-------+ +-------+-------+-------+
| . . . | 8 . 6 | 1 . 4 | | 3 2 5 | 8 7 6 | 1 9 4 |
| . . . | . . 4 | 3 . 2 | | 8 9 6 | 1 5 4 | 3 7 2 |
| 7 . . | . . . | . 8 . | | 7 4 1 | 3 9 2 | 5 8 6 |
+-------+-------+-------+ +-------+-------+-------+

45.......6.79.5....89.......64.87..5...6..4..97.5........8.6174.....43.27......8.
451378269627915843389462751164287935538691427972543618295836174816754392743129586
45.......6.79.5....89.......64.87..5...6.94..97.5........8.61.4.....43.27......8.
453728619617935248289461753164287935538619427972543861325876194896154372741392586

pushed me to stay on the safe side and to keep the full power of the vicinity.

[champagne]

Is your vicinity search of -1/+1
A. kick 1 cell out and replace that cell with another given? ( I used this in the Patterns Game) or
B. kick 1 cell out and any other empty cell can be the cell with some given?

I did not yet answer to your B. point.

When I searched new potential hardest puzzles, I did this in blind mode. Try all possibilities, check if you get a valid and minimal puzzle.

This remains the basis here, but with the help of unavoidable sets.

When you kill a cell in a minimal puzzle, you have a multiple solutions puzzle.
If you use gridcheker, you get in stdout a list of solution grids
If you us my live brute force call, you get, in the limit of 50, the corresponding unavoidable sets having no subset.

Then the problem can be split in 2 different cases.

If you want a new puzzle in the same solution grid, the new cell must hit all these UAs
If you want a new puzzle in another solution grid, you can try each of the solution grids given by gridcheker ( or by the corresponding list of UAs).
But you have to see what list of UAs appears in the selected solution grid whith the "puzzle -1".

[coloin]

I first worked on a similar idea but in a general situation, I don't see which cells should be ignored

The cells that are ignored [ in the example] are the 1/2/3 in Box 9 .. plus dont replace 1/2/3 into any of B12345678
I guess you could separately move the 1/2/3 clues around in Box 9

Not having a means to implement this I was able to purge the grids which were not [xxxxxx110} or {xxxxxx111} - it was slow but it did work

[champagne]

I first worked on a similar idea but in a general situation, I don't see which cells should be ignored

The cells that are ignored [ in the example] are the 1/2/3 in Box 9 .. plus dont replace 1/2/3 into any of B12345678
I guess you could separately move the 1/2/3 clues around in Box 9

Not having a means to implement this I was able to purge the grids which were not [xxxxxx110} or {xxxxxx111} - it was slow but it did work

Surely, I can see which cells to lock if I have a tridagon pattern in the solution grid and if i want to keep it.
But we know that many puzzles in mith's file don't have the tridagon pattern in it.

On the other side, with the constraint to have only minimal puzzles in output, my vicinity search is relatively slow.

I keep the door open for the first cycles, but this remains a point to see later

[coloin]

..... tridagon pattern in it.

Thinking about cells to keep in the solution grid ....

According to denis the "tridagon pattern" would be the impossible pattern of 1/2/3 in box 5 which isnt by definition in the solution grid
If you have/keep a diagonal pattern of 1/2/3 in box 1,2 and 4 in the solution grid [ this is always be morphable to the canonical version].....then
then the valid triple 1/2/3 pattern in box 5 wont be a "tridagon" ....

it can be a valid diagonal triple as in recent "no-tridagon" puzzles

Code: Select all

..x
.x.
x..

or more usually in many of mith's puzzles it will be one of

Code: Select all

..x
..x
.x.
or
..x
.xx
...

by and large if you keep the 1/2/3 diagonal triple in box1,2 and 4 .. I think box 5 sorts itself out
and of course you want to keep 2 or 3 of 1/2/3 somewhere in box 9

[champagne]

Hi coloin,
some comments on this topic.
________
In this experimental phase, I would not miss cases as the example given by “blue” copied in recent posts.
_________
Finding such cells is not easy if you don't know where is the magic square and we know that some solution grids don't have it, what makes the job tougher.
_________
To avoid redundancy, we must use a canonical morph of the puzzle. In my process, I use the Solution grid rank + puzzle 14 Byte bit field.
This means that most of the vicinity search will be done with this start, far from the morph that you describe.

=================

I am working on the results of my first big test. I have several bugs to fix , possibly some of them in fresh DLLs but I have now a reasonable view of what will be the process.

I started from 58 expanded puzzles rating 11.5 11.6
This is 468 minimal puzzles giving a little more than 60000 puzzles assumed to be minimal and passing the filter of easy puzzles.
To clear redundancy and filter old known puzzles, I switched to the canonical morph and checked the results with mith's file.
And the new puzzles (nearly all of them here) where rated with SKFR.

I'll comment separately later on the results, Interesting point here is that in this process
naming + redundancy check + rating is as long as the vicinity search.

IMO, This does not make a priority for a reduction of the cells to use in the vicinity process

[coloin]

of course ... some puzzles will be missed... except the vicinity process would be speeded up...
see this puzzle senbazuru

Code: Select all

+-------+-------+-------+
| 9 8 . | . 7 6 | . . . |
| 5 . 6 | 8 . 4 | 7 . . |
| . 7 4 | 9 5 . | 8 . . |
+-------+-------+-------+
| . 9 8 | . . . | 6 . . |
| 7 . 5 | . . . | . . . |
| 6 4 . | . . 8 | . . . |
+-------+-------+-------+
| . 6 9 | 4 . . | . 7 8 |
| . . . | . . . | 9 . 3 |
| . . . | . . . | 4 2 . |
+-------+-------+-------+

a quick way to generate potential hard puzzles would be to just exchange clues in B3B5B6B7B8 but not using 1/2/3
and i think I remember remarking that the number of ED B1B2B4 isnt that great either !!!
will process your published high raters to see if they are in my list.. !!

[champagne]

seen by my vicinity search, your puzzle (skfr rating 10.1)has 4 minimal
I get in 3m30s 325 "minimal" puzzles passing the filter for easy puzzles

I have now to clear redundancy and to rate this to see what are the high ratings

[champagne]

Final result on my side
255 different puzzles

and my best SKFR ratings

Code: Select all

....76...5..8.4.....495.8...98...6..7.5......64...8.5...94...78........3....6.42.;10.1/1.2/1.2
....76...5..8.4.....495.8...98...6..7.56.....64...8.5..694...78........3......42.;10.1/1.2/1.2
....76...5..8.4.....495.8...985..6..7.5......64...8....694...7.........3....6.42.;10.1/1.2/1.2
....76...5..8.4....7495.8...98......7.5...3..64...8....694....8........3......42.;10.1/1.2/1.2
....76...5..8.47....495.8...98...6..7.5......64...8....694....8......9.3......42.;10.1/1.2/1.2
....76...5..8.47....495.8...98...6..7.5......64...9....694....8......4.3......92.;10.1/1.2/1.2
....76...5..8.47....495.8...98...6..7.5....8.64........694....8....8.9.3......42.;10.1/1.2/1.2
....76...5..8.47....495.8...98...6..7.5.4....64...8.5..69.....8........3......42.;10.1/1.2/1.2
....76...5..8.47....495.8...985..6..7.5....8.64...8....694.............3......42.;10.1/1.2/1.2
....76...5.68.4.....495.8...98...6..7.5.......4...8.5..694...78........3....6.42.;10.1/1.2/1.2
....76...5.68.4.....495.8...98...6..7.56......4...8.5..694...78........3......42.;10.1/1.2/1.2
....76...5.68.4.....495.8...985..6..7.5.......4...8....694...7.........3....6.42.;10.1/1.2/1.2
....76...5.68.47....495.8...98...6..7.5.......4...8....694....8......9.3......42.;10.1/1.2/1.2
....76...5.68.47....495.8...98...6..7.5.......4...9....694....8......4.3......92.;10.1/1.2/1.2
....76...5.68.47....495.8...98...6..7.5....8..4........694....8....8.9.3......42.;10.1/1.2/1.2
....76...5.68.47....495.8...98...6..7.5.4.....4...8.5..69.....8........3......42.;10.1/1.2/1.2
....76...5.68.47....495.8...985..6..7.5....8..4...8....694.............3......42.;10.1/1.2/1.2
.7..56...5..8.4....8497.5...98......7.5...3..64...8....694....8........3......42.;10.1/1.2/1.2
.8..76...5....47....495.8...98...6..7.5......64...8....694....8....8.9.3......42.;10.1/1.2/1.2
.8..76...5..8.4....7495.....98......7.5...3..64...8....694....8........3......42.;10.1/1.2/1.2
.8..76...5..8.47....495.....98...6..7.5....8.64...8....694....8......9.3......42.;10.1/1.2/1.2
.8..76...5..8.47....495.8...9....6..7.5......64...8....694....8......9.38.....42.;10.1/1.2/1.2
.8..76...5.6..47....495.8...98...6..7.5.......4...8....694....8....8.9.3......42.;10.1/1.2/1.2
.8..76...5.68.47....495.....98...6..7.5....8..4...8....694....8......9.3......42.;10.1/1.2/1.2
.8..76...5.68.47....495.8...9....6..7.5.......4...8....694....8......9.38.....42.;10.1/1.2/1.2
....74...5.48.6....7695.8...985..6..7.5.......4...8....694...78......9.3......42.;10.2/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...8.5..694....8......9.3......42.;10.2/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...8.5..694...78......9........421;10.2/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...8.5..694...78......9........423;10.2/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5.4....64...8.5..69....78........3......42.;10.2/1.2/1.2
....76...5..8.47....495.....9....6..7.5....8.64...8....694...........9.38.....42.;10.2/1.2/1.2
....76...5..8.47....495.8...98...6..7.5......64...8....694...78......93.......42.;10.2/1.2/1.2
....76...5.68.4....7.95.8...98...6..7.5.......4.7.8....694...784.....9.3......42.;10.2/1.2/1.2
....76...5.68.4....7.95.8...98...6.77.5.......4...8....694...784.....9.3......42.;10.2/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8.5...94...68......9.3......42.;10.2/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8.5..694....8......9.3......42.;10.2/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8.5..694...78......9........421;10.2/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8.5..694...78......9........423;10.2/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.4.....4...8.5..69....78........3......42.;10.2/1.2/1.2
....76...5.68.47....495.....9....6..7.5....8..4...8....694...........9.38.....42.;10.2/1.2/1.2
....76...5.68.47....495.8...98...6..7.5.......4...8....694...78......93.......42.;10.2/1.2/1.2
....76.....68.4....7495.8...985..6..7.5.......4...8....594...68......9.3......42.;10.3/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...8....6.4...78.....79.3.....942.;10.3/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...8....694...7.....8.9.38.....42.;10.3/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...8....694...78......9.3......42.;10.3/1.2/1.2
....76...5..8.4....7495.8...98...6..7.5......64...9....694...78......4.3......92.;10.3/1.2/1.2
....76...5..8.4....7495.8...985..6..7.5......64..98....694...7.........3......42.;10.3/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8....6.4...78.....79.3.....942.;10.3/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8....694...7.....8.9.38.....42.;10.3/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...8....694...78......9.3......42.;10.3/1.2/1.2
....76...5.68.4....7495.8...98...6..7.5.......4...9....694...78......4.3......92.;10.3/1.2/1.2
....76...5.68.4....7495.8...985..6..7.5.......4..98....694...7.........3......42.;10.3/1.2/1.2
.8..76...5....4....7495.8...98...6..7.5......64...8....694...78....8.9.3......42.;10.3/1.2/1.2
.8..76...5..8.4....7495.....98...6..7.5....8.64...8....694...78......9.3......42.;10.3/1.2/1.2
.8..76...5..8.4....7495.8...9....6..7.5......64...8....694...78......9.38.....42.;10.3/1.2/1.2
.8..76...5.6..4....7495.8...98...6..7.5.......4...8....694...78....8.9.3......42.;10.3/1.2/1.2
.8..76...5.68.4....7495.....98...6..7.5....8..4...8....694...78......9.3......42.;10.3/1.2/1.2
.8..76...5.68.4....7495.8...9....6..7.5.......4...8....694...78......9.38.....42.;10.3/1.2/1.2
....76.....68.4....7495.8...98...6..7.5.......4.7.8....594...68......9.3......42.;10.4/1.2/1.2
....76...5..8.4.....495.8...98...6..7.5......64...8....694...78......9.3......4.2;10.4/1.2/1.2
....76...5..8.4....7495.....9....6..7.5....8.64...8....694...7.......9.38.....42.;10.4/1.2/1.2
....76...5..8.4....7495.8...985..6..7.5......64...8....694....8......9.3......42.;10.4/1.2/1.2
....76...5..8.4....7495.8...985..6..7.5......64...8....694...78........3......42.;10.4/1.2/1.2
....76...5.68.4.....495.8...98...6..7.5.......4...8....694...78......9.3......4.2;10.4/1.2/1.2
....76...5.68.4....7495.....9....6..7.5....8..4...8....694...7.......9.38.....42.;10.4/1.2/1.2
....76...5.68.4....7495.8...985..6..7.5.......4...8....694....8......9.3......42.;10.4/1.2/1.2
....76...5.68.4....7495.8...985..6..7.5.......4...8....694...78........3......42.;10.4/1.2/1.2


[champagne]

My first "big test" was on the direct search output for the first 20 million solution grids;
I tested all puzzles in output rating (SKFR) 11.5 and 11.6.
I have seen 2 bugs,in the run,
one (a stupid one) in my naming process forcing me to rebuild mith's file names
one in the minimal builder DLL, partially fixed. I hope to finish the fix in one or 2 days.

This was 58 expanded puzzles giving 468 minimal.
I got in output 61250 puzzles "minimal passing the easy filter"
This in 1118 solution grids.

No hit seen in mith's file (subject to ...) and at the end 6493 ratings >= 10.5
but 42441 puzzles rating >=10

This is a ratio output >=10 / "minimal in input" around 100.

If such a ratio is realistic for other runs, We can have quickly a huge file to share.
So far, I think of a new file in my google drive but I don't know yet what to put in it.

For sure, the puzzles would be shown with the solution in minlex morph like these ones

Hidden Text: Show

Code: Select all

12...6..94.6.89....8.12.4....4.....8......5......4.3....169....86...2...9.2..4.1.;11.6/1.2/1.2
12...67...5...9...7.912.......6.58.4.........965...2...725.........71...6.19.....;11.6/1.2/1.2
12...67...5.7.9.....912.......6.58.4.........965...2...725.........71...6.19.....;11.6/1.2/1.2
12...67...56....2.7.9.....52..91.6.7......29........51.....59.....397.....784....;11.6/1.2/1.2
12...67...56....2.7.9.....52..91.6.7......29........51.....597....39......784....;11.6/1.2/1.2
12...67...56..9.2.7.9.....52..91.6.7......29........51.....59.....3.7.....784....;11.6/1.2/1.2
12...67...56..9.2.7.9.....52..91.6.7......29........51.....597....3.......784....;11.6/1.2/1.2
12..56....5.7.9.....9.2............8...617....67...3...9...25.7.........67.5.19.2;11.6/1.2/1.2
12..56....5.7.9.....9.2............8...617....67...3...9...25.7...9.....67.5.1..2;11.6/1.2/1.2
12..56....5.7.9.2...9..............8...617....67...3...9...25.7.........67.5.19.2;11.6/1.2/1.2
12..56....567.9.....9.2............8...617....67...3...9..625.7..........7.5.19.2;11.6/1.2/1.2
12..56....567.9.....9.2............8...617....67...3...9..625.7...9......7.5.1..2;11.6/1.2/1.2
.2.4.6.....6.8.1....91.........9..37..8....4....6.2.5..429.....6.12.8..489.......;11.7/1.2/1.2
.2.4.6.....6.8.1....91.........9..37..8....4....6.2.5..429.....6.12.8.948........;11.7/1.2/1.2
.2.4.6.....6.8.1....91..4......9..37..8.........6.2.5..429.....6.12.8..489.......;11.7/1.2/1.2
.2.4.6.....6.8.1....91..4......9..37..8.........6.2.5..429.....6.12.8.948........;11.7/1.2/1.2
.2.4.6.....6.891....91.........9..37..8....4....6.2.5..42......6.12.8..489.......;11.7/1.2/1.2
.2.4.6.....6.891....91..4......9..37..8.........6.2.5..42......6.12.8..489.......;11.7/1.2/1.2
.2.4.6...4.6.8.1....91.........9..37..8.........6.2.5..429.....6.12.8.948........;11.7/1.2/1.2
.2.4.6...4.6.8.1....91.........9..37..8.........6.285..429.....6.12...948........;11.7/1.2/1.2
.2.4.6...4.6.891....91.........9..37..8.........6.2.5..42......6.12.8.9489.......;11.7/1.2/1.2
.2.4.6...4.6.891....91.........9..37..8.........6.285..42......6.12...9489.......;11.7/1.2/1.2


I can also release a file containing the puzzles in my preferred storing mode in this work :
solution grid rank
binary pattern of puzzle in the solution grid

I have to fix my "minimal builder" DLL, but as the current version misses a small number of minimal, I started a bigger test, always on the first 20 million solution grids.
This will be on 699 expanded puzzles including the 58 of the first test, all puzzles rating >=11.2