nickbaxter

The Philadelphia Inquirer is again hosting the National Sudoku Championship, on Saturday Oct 24. 

The organizers just announced a block of rooms at the "official" hotel: Philadelphia Downtown Marriott.

The group code is "sdksdka". Rooms are $99, I think just for Friday and Saturday nights. Use this link for Marriott online reservations.

The results of the 2009 Google U.S. Puzzle Championship are now posted on the US Team site. And for the forth year in a row,motris had the top score. For more on the results, check out the championship web site.

Here I would like to give insight on only the test and the puzzles.

Battleships
Puzzle #1 has always been an easy puzzle, designed to fill time while the rest of the test is printing. I finally figured out that adding white space will cause the page to print faster, so I moved puzzle #2 to the second page. This (and the very tight formatting used at WSC4) also got me thinking about the overall pagination of the test. For the hardest problems (20-23), I separated them one per page to allow more space for notes, similar to a typical WPC one-puzzle-per-page format.

Nikoli
I have been in contact with Nikoli staff since 2000, but never seriously considered approaching them to contribute to the USPC until this year. Instead, in 2002 I sent email to some of my regular contributors, prompting them to write their own versions of Masyu, Yajilin, and Corral. After a successful partnership with Nikoli to provide the 2008 US Sudoku Championship puzzles, the time was finally right to get them involved with the USPC. My hope is that resulting cross-promotion will increase English-language participation at Nikoli.com, and induce implementation of an English language Nikoli forum.

(I had occasion to visit the Nioki office in Tokyo in December. Nearby Ryogoku subway station does in fact have Sudoku-inspired artwork!)

I wanted the Sudoku puzzle to be a modest #2 puzzle. But I wanted the others to be more challenging. To get the right level, Yajilin and Masyu needed to be Giant sized. Some don't like the size, but in order to get the "Nikoli solving experience", I think you'll just have to get used to it!

KenKen®
This Sudoku variation has become surprisingly popular. canadianpuzzler submitted an interesting diagramless KenKen variation (that should appear on LJ soon), and I felt compelled to include a standard version as a baseline. Test solving revealed a slight flaw in Craig's puzzle, and I decided not to use it, replacing it with much harder puzzle.

I decided to replace the generic KenKen with something a little more interesting; fortunately Bob Fuhrer at Nextoy anticipated my request with a page full of operationless puzzles from which I got to choose the best.

Writer's Block
To better ensure that a simple one-word answer wouldn't get cooked, I asked canadianpuzzler to add three more non-answer words. I wanted JD SALINGER to be the extra 8-letter answer, but the initials created an unreasonable ambiguity.

Coordinate Pairs
This had a similar feel to the Dot Triangles from last year, but certainly one fewer dimension made the overall combinations a bit easier to handle. Still, this is still pretty much a trial-and-error type puzzle, and might not have much of a future on the USPC.

Triangular Skyscrapers
I spotted this at the 24 Marathon, and asked Aziz to make one for this. Anyone who has solved this knows that the triangle orientation constraint is key, and the more efficiently you can use this along with the Skyscraper rules, the faster you will solve it. The original version included a piece inventory, similar to what you would see in the Domino puzzle. But this really wasn't helpful as is. More useful (at least to me) was grouping the inventory by which numbers were used for each orientation. I decided not to include any piece inventory diagrams, so that solvers would have to discover for themselves what would be most useful.

Window Pain
I suppose the name says it all! This was inspired by a design I saw while reviewing Scott Kim's 2010 Page-a-Day Calendar. motris was the first to sniff out my mock embarrassment in creating such an obnoxious puzzle; devjoe triggered an amusing thread. But experienced solvers should know by now that most counting puzzles on the USPC have some reasonable counting-without-counting technique. The official statistics show that at least 50% of the solvers still haven't figured out that "counting" is rarely a good technique.

The intended technique was to look at the spacing of the vertical and horizontal lines separately, tabulating how many gaps of 1-11 there were in each direction, then multiplying corresponding values and adding them up. Not too bad.

onigame used an alternate approach, with slightly more work, but again not requiring that each individual square be counted. He looked at the upper-left to lower-right diagonals. All squares must have exactly two vertices on one of these 23 diagonals. So for each diagonal, count the number of lattice crossings, including points on the edge (the diagonal starting at the upper-left of the grid has 6 such points), and compute n(n-1)/2. Sum all these values to get the answer!

Masyu
This year's answer key format was a step backwards. Next time I'll go to a fixed set of node that the solver has to put into the correct order. And I'll try to use this or similar techniques for similar route puzzles (but probably not Fences and Corral).

Lucky Sevens
The initial version of this puzzle had FUCHSIA as the key unused word. Fortunately one of the test solvers had searched for all words that could cross seven other words. Not only did FUCHSIA not satisfy this constraint, but because of the C he quickly discovered that it could not possibly appear anywhere in the grid! Shawn replaced it with TERRACE, with was much better.

Sweet Sixteen
I was surprised that only 39 people even attempted an answer this problem; the points shouldn't have been intimidating; perhaps the late position was. It actually wasn't much trouble. The sum of 2-16 is 135, so the bottom three rows total 117, and the common row sum is 39. The corner 2-edge triangles must all have the same total, 18, which gives 81 as the sum of the central 3-edge triangle (and forcing the values 11-16). Using this and the row sum, you finally find that the central 2-edge triangle must contain 14-16, and the rest is easy.

I named the puzzle to suggest a square arrangement (instead of triangular) to anyone trying to plan ahead. And yes, I was tempted to make this a 16 point problem. But then I thought better of it for fear of even the smallest chance that the rankings could be impacted by such a whimsical change.

Di-Agony
Another surprise: only four people solved this. It's really a very pretty puzzle, but easy to break if you're not careful.

Four Square
Another design I found on the 24-Hour Marathon. I liked the way the quadrants interacted, and that you had to continually shift focus to make progress. I think this one received the most complements from the solvers.

Inside/Outside Corral
This was not in the initial play test. The first test took about the same time as last year. And since there was no nasty wild card (i.e. Point Triangle), I decided that one more tough puzzle was needed. I had this one waiting for three years for just such an occasion, and I think if finished off the test perfectly.



Overall Composition
Overall the test had the desired balance between puzzle types. Magic Puzzle'Rs was overvalued (it really should have been 15 points), so Lucky Sevens was the only bona fide non-logic grid worth over 15 points. This was not intentional, since I like to have a tough word puzzle as one of the final two. Just unlucky this year, I suppose.

Web Site Improvement
I've hired an MIT EECS major as a summer intern to write a new answer entry page. I hope to have an answer page that accepts one answer at a time, with format validation, a clear button, and have the answer page refresh with the previously submitted answers, giving immediate confirmation that the answers have been received.
 

The official test page, with file downloads, is now up.

This year I decided to include the names of the puzzle contributors in the test, rather than wait until afterwards to give credit where credit is due. Cynical test solvers have already suggested that I'm doing this at the request of one of the publishers newly represented this year. Not true!

I've actually wanted to do this for a while. Historically, I did not include names for fear that solvers would use this information to gain some sort of advantage. Ten years ago, that might have been true. But the list of designers that I use is now pretty consistent. And in recent years there is much better world-wide sharing of puzzles, national and practice tests. The active solvers are likely to know the designers anyhow. So by pre-announcing, the playing field is again level for those curious to do further research on the designers.

There are some possible spoilers with this information. For example, knowing that a Masyu puzzle is from Nikoli, rather than Dave Tuller, might tell you something about your solving experience and how you might approach the puzzle--since one is assuredly hand-constructed while the other is likely computer-generated. That is an interesting theoretical discussion, and I look forward to hearing people's thoughts beforehand, and reaction afterwards.

I hope to be a little more chatty this year; I see that I'm picking up this blog after two years of inactivity. Last year I went on vacation the day after the test, and did most of the scoring from a wi-fi enabled cantina near Cozumel. The scoring might be disrupted again this year if I attempt to relive past WSOP glory in Vegas on Monday. Catching up on sleep would be a much better choice.

The 2007 Google U.S. Puzzle Championship was yesterday, June 16. I started this journal to establish a focused a forum for feedback on the test and puzzles.

Administratively, this is the cleanest test I've run so far--as far as I know there were zero technical glitches. I finally invested in some software that allowed me to do stress/capacity testing of the servers. ThinkFun put us on a new and bigger server which also helped a lot. In January I ran a small test for the Sudoku Team qualifying, and the new server failed to handle just 25 people hitting the password file. It turns out that just minimal demand caused the server to cache the non-updated password file! After getting that finally fixed and tested last month, I'm very happy with our new home. Also, I rewrote the timer page so that it automatically updated itself on the client (pinging the server only every 15 minutes), which I believe cut down on the pre-test panic traffic.

As for the test, I did have a feeling that a couple people could finish the test this year. As someone pointed out, there were fewer puzzles in the 15-25 point range this year, and more in the 30+ range. Obviously there is more variance in the latter range, and thus more likely that someone could get hot and finish the test.

Based on preliminary scoring, motris and onigame proved my premonition to be correct. Wei-Hwa was very cooperative and submitted an answer after every problem, giving me the same data that I get from the test solvers. Thomas was not so cooperative, submitting just three virtually identical and complete answer sheets (each with a different interpretation of the Circuit Maze answer key). His first entry was logged at 2:06:21. I'll reserve comment on that time until the drug tests and the rest of the investigations are completed!

Now on to the puzzles. As always, I was really pleased with the overall quality and variety of the puzzles I got this year. I was also pleased to add a couple new designers ( zotmeister and Mark Steere) into the mix. I'll comment here on puzzles that are noteworthy either in my mind or based on Journal comments already posted.

Battleships and Diagonal Sudoku
I always put familiar designs on the first page so that people have something reasonable to do while the rest of the test is printing out. There are supposed to be routine; rpipuzzleguy has no excuses here.

Count Me In
Thomas half-jokingly claimed this is not a puzzle ("I counted 19 excellent puzzles on the USPC. Maybe my count is one low, but that is a problem I have!"). I selected this puzzle because it had a very interesting quality that I had hoped other solvers would discover and take advantage of, as jrivet did. Other than the vertices of the outside triangle, every point of intersection contained the right angle of exactly two triangles. So the problem reduces to simply counting the interior points of intersection and remembering to count the outer triangle.

Comic Strips
I've received mixed reviews on this (mostly positive), and remain very happy with how I presented it. One reviewer didn't like the trick and asked for an explicit hint (I was considering "There is something funny going on here!"); and on the other hand one reviewer wanted me to add an extraneous 8x8 grid, to make the puzzle really evil. jdyer sent me email the night before the test asking if the grid would be 8x8 or if the strips could be offset. I had to be honest in my answer but also not give an unfair clue; I congratulated him for finding a loophole in the instructions and flippantly said he could arrange the strips anyway he liked.

Circuit Maze
Andrea Gilbert suggested that I consider Mark's design. After looking at the original, I requested a version that was about 1/10 the size, to make it easier and so I could enlarge it to a reasonable size. If anyone thought this one was too easy, I suggest you try the original!

Arrow Ring
I like this style, but am very dissatisfied with the various answer formats I've tried. The inside/outside counts was another failure, and I'm not sure why I didn't figure this out during the test solving. Suggestions are welcome. This particular puzzle was painfully easy. I cooked up a replacement, but it was too late to use. Expect something more challenging next year.

Masyu
This is my all-time favorite Nikoli design and was happy that Dave Tuller finally made one for me. It's the one style where I really get a sense of the designer's intent. This puzzle is different (I suspect it's computer generated), and as jdyer points out, it requires a bit of hunting or extreme intuition to find the right starting point (the black between E and I). The black between the M and N uses logic that I've found in Nikoli puzzles, but that wasn't enough to get started.

Fence Posts, Fridge Magnets, Corral, Double Murder, ORu Kakuro
These puzzles were the heart of the test. Most were good, enjoyable variations on common styles, but otherwise I don't have much to say about them. Corral is another favorite of mine that doesn't appear often enough in Nikoli, and I really liked the logic around the perimeter.