Greek Sudoku and Puzzle Championship writeup

This post ended up as a wall of text, so look under the fold at your own peril.

TL;DR version

I did much better at sudoku and much worse at puzzles than I anticipated. Puzzles were of modest difficulty but rather neat. Test format leaves a lot of room for improvement, including typesetting, problem type selection, and examination format. Still, fun overall.

I can hear some of you foreigners reading the title and going "There are Greek Sudoku and Puzzle Championships? Cool - where can I get the puzzles?" The answer is likely nowhere. The contest was held online via email, which proved exactly as kludgy as it sounds. Since there was neither licensing nor author / editor information available, I don't feel I can publish problems or the files, nor can I provide them upon request. You may try your luck here if you must.

I can also hear some of you locals reading the title and going "There are Greek Sudoku and Puzzle championships? Why am I only finding out about this now?" It is likely because
  1. you respect your privacy enough not to belong to Facebook,
  2. you ignore the existence of the Facebook group.
  3. you are not a member of the Greek Mensa chapter.
  4. your grapevine has no-one in the loop.
I barely failed the fourth criterion, so I found out about it somewhat belatedly, and took both tests. In a nutshell, I am rather satisfied at my Sudoku performance, and very disappointed at my puzzle performance. Given that a lot of criticism of the actual tests follows, one may think this amount of hopefully constructive (but perhaps withering) commentary is bleeding of performance dissatisfaction onto the test itself. I beg to differ.

On the test document format 

 Let's start with the language. These tests are set out by a Greek organisation, and test exclusively Greeks, in Greece. So of course, they are written exclusively in English. It has been explained to me this is because the event also serves as WSC/WPC qualifier, whose instructions are in English. Thus, I can only assume this has been a team problem in the past, even though the stereotypically less than fluent Japanese teams seem to do fine with just the on- and off-line forums that go over the rules and examples. Also, this additional barrier to entry flies in the face of one of the two stated current goals of the organisation, id est popularising logic puzzles in Greece (the other being, in case you're wondering, "can we be second to last for once at the WPC? Please?"). And yes, I appreciate the irony of having this discussion on the topic of language-neutral puzzles. Finally, to add injury to insult, WPC solution keying was incoherent until you figured out that line actually means row, while row actually means column.

Second, typesetting was a little dense for an important competition, with GSC clocking in 2 to 4 problems per page, and GPC 3 to 5. For the standard Sudoku of GSC and the Kakuro of GPC, this meant that note-taking could get cramped when your pencil got blunt. Granted, maybe it's just the combination of my large and ungainly hand writing and horrible eyesight that make me crave larger grids.

Wrt to the GPC however, a lot of other critical information seems to have been squeezed out onto the instruction booklet in the process. No examples, which I can live without. No allowance for key extraction either, which was a first for me and proved very ungainly. Keys were described only on the final page of the instruction booklet, which the wise had printed out beforehand, while the rest looked at the situation startled and askance. And when I say no allowance, I mean it. Want to know which rows (meaning of course, columns) are part of the key? No namby-pamby arrows for you. Want to know the submarine coordinates? Write your own damn scales along the sides of the grid first. I needn't explain how problematic, for both solver and proctor, this state of affairs is. By contrast, the GSC had one of the most beautiful systems I've seen: for all puzzles, you had to provide the second row, then the second column. And since you can freely fiddle with a sudoku until the final bits fall on the keys, you needn't sacrifice grid completeness for key uniformity (some variants excepted).

Each contest also included two cover pages, which are pointless for an online competition. They were also grey, perhaps for the purpose of punishing those who don't carefully set up their print jobs out of maximum time and ink. For the GPC, they were the only place where you could find out the problem value. I find that scores next to the puzzles make puzzle selection much easier, and I suspect I'm in the majority on this.

On the Greek Sudoku championship

Déjà vu
The test lasted 140 minutes and added to 500 points. You had to solve 12 standard sudoku worth either 11,17 or 27 points each for a partial total of 220 points, and 8 variant sudoku worth 11 to 49 points each for a partial total of 280 points.

I expect a Jan Mrzowski level solver could finish this test in about an hour (thus scoring about 740 points with the time bonus). I expect to have scored 384, which is so much better than the 250 I predicted. This was partly me having a freakishly good day and spending most of the time in "the zone", and partly figuring out very early on that the variants were heavily favoured by the scoring.

Grading of the standard Sudoku and the range of difficulties felt particularly nice. I can't really be impressed by any standard Sudoku these days, but the grids looked and solved pretty enough.

The variant section was another story, with few of the problems making a technical, experiential or aesthetic impression on me, and scores which correlated very poorly to the time I needed on each. First of all, are we still considering diagonal and irregular Sudoku variant? We are? Oh, OK.

Second, the variant selection felt lacklustre, mostly at the type level than the problem level.
It included two 6x6 problems (a cute chain sudoku and a why bother irregular), then standard size
  • diagonal: meaty if somewhat bland.
  • even: aka the world's least interesting variant type.
  • killer: OK, I liked this one even though it took me far longer than it should.
  • consecutive: Blinding break in, followed by some token nastiness when you think you're off to the races.
  • quadruple (some two by two regions have all their values given out of order): See above. Worth far too much.
  • frame: To paraphrase Janis Joplin "It's all the same f*ing problem man, it's all the same f*ing problem." - they all solve the same, more so than even standard sudoku.

In my ideal world, the test would be more encyclopaedic in its type coverage. I'd have liked it to contain at least one, and possibly two, meaty variant geometry problems, rather than the two 6x6 apologies. Which would probably have to be an irregular and a wackily-shaped grid (toroidal, tablecloth, rhomboid, Penrose or something). Grids with non-standard fill elements should probably make an appearance, by which I mean not letters instead of numbers, but things like surplus/deficit. An additional extra region problem to complement the diagonal would be nice (say Windoku, Colour, or Renban), and so would a second arithmetic problem (like Arrow or Kendoku). What I could use less of are pencilmark variants - they change your thought process very little, and putting three high scoring ones on a test (even, quadruple, frame) feels a bit much.

On the Greek puzzle championship

Bad, Neotubby, bad! (Star Battle)

The test lasted 120 minutes and added to 535 points. You had to solve 12 standard puzzle types worth 10 to 60 points each for a partial total of 360 points, and 2 make your own crossword style puzzles worth 80 and 95 points each for a partial total of 175 points.

I expect an Ulrich Voigt level solver could finish this test in about 45 minutes (thus scoring about 760 points with the time bonus). I expect to have scored 360, which is so much worse than the 440-540 I know am capable of. This was partly me being completely out of it by running about all day in the scorching heat, and partly having every thing that could go wrong go wrong (a computer crash, a paper jam, a combined water and ink spill and a persistent wrong number caller, for your schadenfroh pleasure).

Those who have been paying attention so far may have already figured out that the regular puzzle grids need have been pretty small to fit 5 to a page. They were also easy, almost uniformly far too easy for the stated purpose of selecting WPC representation.
They were in order:
  • Battleships - 42: Battleships problems can be made to work beautifully, but are all too commonly like this, exploring poorly prunable search trees. One of the few not too easy problems, which still doesn’t mean it is contest-worthy. Also, instructions reference the "given fleet" which isn't actually described or depicted in any form.
  • Minesweeper - 22: A type with minimal thinking required, especially for problems where the mine total isn't given. Pointless.
  • Easy As ABC - 10
  • Skyscrapers  - 13
  • Kakuro - 30: The best problem of the lot. Good fun, if stingily scored.  Reasonable grid size, varied technique use, with note taking under control.
  • Fences - 24: The most difficult thing about this puzzle is the key mechanics: "Give the amount of 2s that are outside the fence and the amount of 3s that are inside the fence." Not only is this mechanism famously error-prone, but it's topped with the outside/inside cherry for no discernible reason.
  • Nurikabe -19
  • Star Battle - 21: Very fast and generously scored, I appreciate the super-clean construction and the hidden, centred vertical region -column overlap.
  • Tents - 30: Another generous score.
  • Dominos - 26: Domino puzzles can often be nasty in much the same way Battleships problems can. This was almost entirely devoid of such nastiness, and was in fact a bit pleasant, which is about as excited as I can get for Dominos.
  • Build A Maze - 60: Weird in so many ways. Despite the standards claim, it's pretty novel, and I had only done a couple of those in the past. It's that problem where you need to design a 1 cell-thick path from A to B visiting all other cells. Path is forced by on-grid walls, which are described picross-style. This one was additionally streamlined by having the endpoints on opposite corners. I spotted a 1 deep, 2 wide branching point and it was off to the races, with thinking far outpacing my writing just this once. The scoring felt very much off - if it takes you less time than the Easy as ABC, it probably should be scored similarly, no matter how novel the type or pretty the intended path to solution (No groaning, you pun haters). Incidentally, I think this type could really be improved with the addition of some Corral-style clues - it's an idea I'll have to try out eventually.
  • Magnets - 28: Magnets are often a sticking point for me, even though I'm largely not metallic. Still, this would have felt generous at half the points. A key and solving process that meant you needn't draw a single pole weighs in as well.

This brings us neatly to my doom. It's been years since I've done one of those DIY crosswords, so I needed practice. Like most homework, knowing it needs to be done is far from convincing yourself to do it, and I ran out of prep time before I ran out of excuses. I even expected the word list (Adriatic place names). I didn't bother to develop a letter/digram frequency feeling either, which would have been useful. Add to that a dash of greed, which led me to pick the harder problem to focus on for 15 measly points, and what followed was a comedy of almost correct assumptions, also known as errors. Half an inch of rubber later, I've had enough and moved on to the other one. This one had 3 independent, neat break ins, and a bit of "either/or don't matter" logic in the middle. So I accelerate to completion, prepare to submit, and notice I'm a minute late, and thus 16% of the test short. Hence the sourness forewarning.

I'm generally against putting whales in a test (problems that make up too great a percentage of the total score), irrespective of how deserving these problems are of their high score value. That's mostly because they manage to introduce both high individual score variability, and because they tend to lead to score distribution clusters. It also tends to reward a singular skill at the expense of well-roundedness. Putting two identical twin whales in the test just rubbed me wrong, if you'll excuse the pun. The situation was exacerbated by the instruction booklet, which suggested these problems were worth about 20% per expected unit of time spent on them less than the other problems, which can be expected to lead more people to leave them for last.

Again, I'm unsure what the puzzle type selection is meant to achieve. It doesn't represent the trends of the local puzzling community, or at least the Northern part I'm familiar with. It isn't nearly encyclopaedic enough: A singular loop, a single (mock-)path problem, very restricted Japanese styles, no variant grids, no gridless problems, no division (like shikaku, pentomino packing, dissections etc) or connectivity problems (pipes, numberlink, anglers, zipline etc), no real arithmetic problem, no observation, counting and so forth keep it far from achieving wide coverage. Finally, it doesn't really seem particularly well matched to the recent WPC styles: no problems with truly deep logic, no variants and no new types don't really seem to be well-suited to an event that last time around had only one or two rounds of standards, which they were in name only. Consequently, i can't figure out what traits the proctors are trying to promote.

Thank yous

All of this is not to say I don't appreciate the effort that went into the event. I am thankful for the initiative and drive on display, which can only work to the benefit of the local community. I am thankful for the time it took to write problems that solve (by and large) cleanly - I wish I could get more of my own efforts to this level of neatness. I appreciate the effort that went into testing, so as to give us a contest with no broken problems. And, perhaps most impressively, I appreciate all the time sunk into what turned out to be individual proctoring-by-appointment, and in the summer season nonetheless. Finally, I would like to thank the readers of this more than 2500 word post - in this day, this kind of patience is rare.