October 27th, 2007


Kakuro R Kool

So, I posted earlier about making some Cross Sums (Kakuro) puzzles. I got some good feedback from Raven on several of the puzzles I shared in that earlier post. Then, at the beach, I hooked Laura on these puzzles, as well. I might not be able to feed all my friends' addictions, but it seems I can give them new addictions, so that's good.

I'm in the process of fancying up the presentation, including writing front matter (introductions, hints), adding puzzle numbers and page numbers, and a second pass on making sure the solutions are legible. I've made cover art, which is moderately pleasing to the eye.

This morning, I went through the pile of puzzles that I had already generated and checked to see how many of them were duplicates of one another, and I was astounded to find a single duplicate amongst my 4000+ puzzles. I also went through and did a filtering pass to pull out puzzles that were generated with buggy code - some puzzles were just not solvable. (Impossible, even for a computer!) Now, the computer telling me that a puzzle is solvable or not is one thing - people's feedback tends to be interesting and valuable in completely different directions; it's hard for a computer to tell me which puzzles are fun.

Right now, I've got three machines working on generating more 23x31 square puzzles - these are the beasts that I'm proud and a little frightened by. I haven't seen any cross-sums puzzles that large anywhere else, so it's a competitive niche that I've got to myself, but they're only for the really dedicated solver. Make one mistake, you could ruin hours of work. Laura's second Cross Sums puzzle ever was one of these, and she finished it. I haven't even finished one of these.

As I was looking through some (very) rough drafts of the puzzle collection, I noticed several of the 23x31 puzzles sharing the same pattern of black and white squares, which makes some sense - I generate the "grid" as a separate process, then later I try assigning numbers to the grid. If I recall correctly, I only had a few 23x31 grids generated as I was making the puzzles, so they very quickly overlapped. There's not much to be done with overlap on the 3x3 grids - what with rotational symmetry and the rules of the puzzle, there's pretty much just one grid (plus one rotation of that same grid). It gets better as the puzzles get larger - and, indeed, I have plenty of 23x31 grids to use. It's just a matter of waiting for the machinery to kick out enough new puzzles for inclusion in the book.

One last piece of the pipe that I don't have in place is a mechanism for selecting puzzles for inclusion. I want to make sure to include the puzzles that Raven indicated as especially fun, but at this point, the puzzles are selected at random. Making a more deliberate selection process is easy enough, I just have to do it.

A side-effect of that is that I'll probably be able to use that same selection mechanism to record what puzzles were used in this book, in case people can't get enough of these puzzles and want to buy some more.

I think that's enough puzzliness for today - I'll let the computers chew away on making new giant puzzles and tackle the polish and selection tomorrow. I know not everybody is interested in geeky cross-number puzzle stuff, but I know a few people are, and in any case, posting about it gives me that extra bit of accountability to actually get this thing done.