how do you rank them, Ruud?

[lac]

I found the Jan 23 puzzle about as hard as i could tolerate.

after 7 hours, I finally got a 'proof' solution though it is so close to
what you get by just assuming one digit and then see if the
puzzle solves, that it is anything but satisfying.

meanwhile, the puzzle Ruud says is the hardest -- I forget which
one was it, Jan 8 I think, but the archive does not preserve those
taunts ... that one I spent under 1 hour at, which is pretty good
for me. So that means that one was hard, but not exceptionally
so.

So Ruud, what is the hardest one you posted yet, and how do
you determine their rankings?

Laura

(Faster ways to do the jan 23rd also most welcome.)

[Ruud]

Hi Laura,

that is an interesting question. For this, I must describe my Nightmare production process, which is so secret, that I'd have to kill you after telling you.

This is how it goes:

1. A random Sudoku generator in C# can generate around 100.000 sudokus a night, depending on the clue limits and symmetry requirements I give it.

2. A preliminary filter removes the sudokus that are solved by singles only.

3. SudoCue evaluates the collections and counts the techniques required to solve it, including occasional requirement of brute force.

4. The puzzles and technique counts are stored in a database. This database has a few predefined queries that pick out the interesting ones.

5. Manual review by me. I check on nice looks, good mixture of techniques, and sometimes load it into SudoCue to check the solving path. If it's OK, I add it to the Nightmare table.

Of the initial 100.000, between 5-10 Nightmares can be harvested, with only 1 or 2 that players really admire.

This is all done a couple of days in advance, so when the new one is available on the website, it is as much a surprise for me as it is for you.

So now prepare to die...

Ruud.

[David Bryant]

(Faster ways to do the jan 23rd also most welcome.)

I sweated over this one for a while before I noticed something. After filling in 15 cells and doing some coloring I arrived at this position.

Code: Select all

34   5    1    7    6    8   34    9    2
 2   6    7   39   39    4    8    5    1
 8  39   349  12    5   12   36   46    7
34   8    2   136  13    9  1567  167  45
 9  34    5  1236   7   123  16   146   8
17  17    6    4    8    5    9    2    3
 6   2   39   58    4   137  157  178  69
 5 1479   8   19    2   17   147   3    6
17 1379  349  58   139   6    2   178  459

There are three corners of a "non-unique rectangle" in r6c1, r9c1, & r6c2. If you make the assumption that the solution is unique you get {3, 9} pairs in column 2 and in the bottom left 3x3 box, and the rest of the puzzle is easily solved.

I haven't hunted down the whole string of inference, but it seems that an argument starting from r3c3 = 4 (r3c3 <> 4 in the solution) might yield a contradiction fairly quickly. Is that any help? dcb