Detailed USPC Post-Mortem

Full 2011 USPC results are now out. As could have been predicted a couple days after the test, I ranked first (415) with Thomas Snyder in second (349). Congratulations also to William Blatt, who made the team with 309 and 3rd, and Dan Katz, who just barely missed the team for the Nth year with 305 and 4th.

Now, what is the rest of this monster post for?

One of the many things I admire about Thomas Snyder is his willingness to share. As probably all of you are aware since he authored the USPC’s official practice test, he freely shares tips, example puzzles, and almost anything else about how he succeeds. I’ve been working pretty hard throughout the past year to patch up some of my weak types, and three or four times I emailed him when I was really struggling to improve on a type and not having much luck with just doing a whole bunch. His responses were always excellent, and he’s probably the main reason I managed to go from terrible at Sudoku to passable. Even at the WPC, I remember at least one instance of another competitor coming to him to talk about a type in the instruction booklet, and he was quite happy to discuss how he planned to approach the puzzle during the contest.

Since I admire that, I’d like to emulate it. Here is a detailed rundown of my 2011 USPC. I’ll be talking about what I did 24 hours in advance after the instructions were released and how the actual 2.5 hours went. If some of you want to know why I didn’t share any of the preparation details in advance, it’s because it would have been awfully presumptive of me and because I didn’t have much time to; it was a busy 24 hours. Next year could be different.

Before I get started with it, I should note that this is not a guide with the best, fastest, or easiest break-ins for each puzzle. They just describe how I personally toppled the puzzles. I’ve found other people mention far more efficient methods than what I used for a few different puzzles, and I’ll often mention those here as a sidenote. So does that mean you should ignore what I did and concentrate on how to find those “best” methods as a means to improve?

Not necessarily, I would argue. My reasons are based off of a point I discussed in my akil oyunlari interview; specifically, see the third to last unbolded paragraph where I talk about heuristics. While some of these mentioned break-ins involve very simple logic and are easy to see through, they may not necessarily be the fastest to find. Remember that your time is not solely based on the steps you use to solve a puzzle; more of it is determined by how much downtime you have between each one where you are figuring out what to do. The important thing is to figure out a method to approach these puzzles that will allow you to reach a break-in as quickly as possible. The simpler the break-in you get the better of course, but that’s not the only thing to keep in mind.

I’ll be doing the rundown in chronological order of my performance (well, approximately, because I wrote the first draft several hours after the competition ended). My usual strategy for doing tests I expect to finish (or get close to finishing) is to go from front to back, skipping what I get stuck on or want to put off, do another pass through or two depending on how much is left, and then pick and choose once I’m down to just a few. So I’ll describe my first pass through, which is where I’ll include all of my notes about my preparation the day before, and then revisit each puzzle in accordance with when I pulled it back out again during the test.

Thanks a ton to Nick Baxter, the organizer, who graciously lent me my time series for this test to help make this rundown more accurate. I’ll be listing a time below each puzzle, which is in general how long since my previous submission it took me to solve and submit a puzzle. This includes a possibly substantial amount of checking time, sometimes a minute or more for a puzzle.

1. BATTLESHIPS

Preparation: None, really. This puzzle rarely poses much of a problem, as any USPC veteran knows.

On the test: I got all the basic steps down, then spent some time staring. Some of these fall prey to simple logic, but I couldn’t see my way through this one. Eventually I plunked down the battleship on one of its few possible locations and saw what happened. Did another bifurcation later and was very fortunate to find I was right on both guesses. It seems this presented problems for others and I’m no battleships expert, so I feel pretty lucky about this one.

Time: 2:29 (0:02:29)

2. SUDOKU

Preparation: I haven’t solved on nikoli.com for the past month. I try to do this before the USPC to get used to solving on paper more and because I think that site’s lenient check button is what made me such a mistake-prone solver (see the second question here for the role nikoli.com played for me in starting to competitively solve). The morning of the test, a couple hours before, I went through the backlog of sudoku puzzles, six or seven of them, with decent solves on all of them. There was only one “error”, but that still scared me to death. I decided I would take things a little slower on this one, taking an extra second to verify that I did compute the right missing set of digits in a row/column/box I was analyzing.

I should note that on LMI and the USPC/WPC last year, Sudoku and any variations were a tremendous weakness of mine. They still are; I’m just semicompetent now so that they no longer wreck an otherwise great performance.

On the test: Like last year, it was pretty tough. Also kind of like last year, it seemed well-themed; at least my solution had a succession of X-wings at the beginning. Even after getting them I struggled quite a bit, but I eventually put it away to my relief. I figured I lost some time on a motris-level solver here, but that’s pretty much impossible for me to avoid, and I was satisfied it wasn’t more.

Time: 4:52 (0:07:21)

3. CROSSED COUNTRIES

Preparation: This was a funny one. I actually was really confused by the instructions. It sounded like a square grid with the answer entry text… what were these gray circles and loops? In the past some word puzzles, like Comic Strips from circa 2005 and Mirror Mirror last year, I actually attempted to solve them under the wrong set of rules and wouldn’t get them, so this made me paranoid and I emailed the USPC people asking if any clarification was possible. The answer said it would be obvious when you saw it; I just decided to trust that. Couldn’t really prepare though.

On the test: Fortunately, it really was obvious when I saw it. It was nice to sort the countries by length; the first thing I try is to focus on the shortest of those lists, which were the 9 and 10 letter ones here. After hitting hard enough on the 10 letter words, looking for what letters were possible in crossings, I eventually broke through and solved it fairly smoothly after that, although it was time-consuming. I solved an entire volume of nikoli Nansuke last year for USPC 2010 practice (these volumes are far harder than many of their others), and it helped me know exactly how to approach these kinds of fill-ins.

Time: 8:47 (0:16:08)

4. BARN STORM

Preparation: I just did a few Spot the Differences puzzles from some past USPCs, particularly ones I skipped in the tests I did for practice. If you read the comments on motris’s blog, you’ll know I was banking on the lack of one last year being a permanent change. It wasn’t to my dismay.

On the test: I put it in the todo pile immediately. My points per minute on these puzzles is abysmal.

5. EASY AS ABCD

Preparation: I was originally planning to copy some out of my (finished) Mensa Math/Logic volumes by Tuller and Rios, but I noticed there were some in a packet of puzzles Wei-Hwa sent out for test-solving in the beginning of the year – they will be part of a to-be-published book. So I just did them. The one on the test was early with a low point value, so I wasn’t too worried even though this is a weaker type of mine.

On the test: The size was compensated for by the amount of givens. This one went down in a pretty standard way. The first placements can be easily gotten by looking for edges where a letter only appears once, which is a perk of having all of the givens. I often get stuck and bifurcate on these, but I didn’t need to this time. I’ve also been trying to reduce how much I do that recently; I may have guessed on this one in the past.

Time: 3:21 (0:19:29)

6. TIMES LIKE THESE

Preparation: As my About page says I browse Art of Problem Solving pretty often for interesting problems to do, and in the 24 hour period before the test I managed to find this post, which had 6 practice puzzles total, although I only did about half. I’ll talk about each as they come up, but one was a Times Like These that I had no trouble with, and after thinking a bit I decided there was no way this type could be hard.

On the test: And I was right. Note a 5 digit given means all the factors are definitely a single digit, so pick three digits, take their product, and see if it’s composed of the other two digits. Some intuition about parity can reduce the brute force nature of this; in the 24678 row you know the number is going to be divisible by 4 at the least, which is a big constraint on the digits of the product. Once you figure the 5 digit givens out, look at the intersecting digit sets for eliminations. This one didn’t take more than a minute or two.

Time: 1:48 (0:21:17)

7. MASYU

Preparation: None. My nikoli.com and crocopuzzle Masyu times, while not the very best, are way more than sufficient. Loop puzzles are always pretty good for me. There are much better ways to use my preparation time.

On the test: There were lots of easy deductions with white circle clusters as well as the black ones to draw in at first. Then in the bottom left, I caught sight of a trick that juno had done in a previous nikoli.com puzzle (still in the Pick of the Puzzles section), involving not cutting off areas with a white circle. I nailed that, then went through a couple more tricks that I think I’ve seen only in his puzzles, like two-possibility black circle chains with one option sealing the loop in. Even if there wasn’t anything truly new, it was still an amazingly well constructed puzzle, and I’m always happy to see those tricks reused.

Time: 4:06 (0:25:23)

8. SQUARE COUNT

Preparation: None actually. I meant to do something at some point, but basically ran out of time to. I never got to motris’s on his practice test either, which would have been very appropriate. The AOPS link mentioned earlier had one, but I skipped it, mostly because there was nothing to check with.

On the test: Relegated to the todo pile immediately. But I did note that it looked quite doable, an impression I did not have about last year’s.

9. CRACK IT ON

Preparation: Above for Easy as ABCD I mentioned getting out a test-solve packet from Wei-Hwa awhile ago. This was actually the puzzle I dug that up for, as there were lots of them there. I did about half of them (the easy ones), but remembered the hard ones taking forever and not being too fun. Since this one was 15 points, I let it go at that. I know motris posted some others from 24HPCs, but I chose to let things be and didn’t do them.

On the test: None of the top left and bottom right blocks were that big, unfortunate because that’s often a quick way to start. I tried analyzing how often first letters were duplicated anyway. Answer: A ton. Spent some time being at a loss, then realized that if the set of first letters was so small, the set of words that could go along the top and left edges can’t be too big. So I listed out the first letters and went through the word list to see which ones only had them and could go on the edge.

When I was done with that, I still had several options for what I could put in those corners. Getting pretty desperate, I played with a few options in the second grid, which had the domino in the top left. I don’t know if I proved it was the only one, but I eventually got putting an E there to work way too well to believe I had gone wrong, and it fell quickly from there.

I later saw others discuss easier break-ins involving rare letters like P or D. To be honest, finding uncommon letters is a method I probably need to get better at employing myself. One reason I avoid it is because you can lose a ton of time if you miss an occurrence of a letter somewhere in the bank, and I often make this mistake.

Time: 4:54 (0:30:17)

10. FLASH CARDS 1

Preparation: I was quite worried about this type; motris had been too if you remember his post about the instructions. I thought it was probably a Friedman type and that it would be like last year’s Hex Equation, which was insanely hard.

So I whipped up a python script to generate these on the fly in an hour or two, and eventually got it to churn out a sheet of 100 of them, some of them utterly trivial, some of them way too long (I was kind of expecting a 10 flash card puzzle for some reason), some of them not unique because I hadn’t coded a checker.

I started going through them and found even the small ones were really tricky. I think I got through about 5 or 6 before deciding that I wasn’t getting too much more out of them; it was just random guessing the whole time. I just crossed my fingers for the ones on the test.

On the test: Four cards! A big relief; most of my practice ones were far bigger. I noted it was actually a Cihan Altay puzzle, so I spent about 10 seconds looking for logic and found none. So I figured I’d try a few guesses and move on if none worked. With little to guide my guessing, I decide to go with 4+35×9 first. So that’s 319 and … 319!? Woo, free 10 points!

Time: 0:55 (0:31:12)

11. FLASH CARDS 2

On the test: The fact that it was one digit divided by two digits caught my eye immediately. So the result is definitely fractional. Moreover, we have to get the same denominator both times, but we divide by one digit in the flipped case. There’s got to be some greatest common divisor in the nonflipped case, and so there’s probably just one way of getting the same denominator. After some time spent, I eventually get a set of three cards that got the denominator to work, and unsurprisingly it gave the right answer. Huge sigh of relief at having solved both of these smoothly.

Time: 1:36 (0:32:48)

12. TAKEOUT

Preparation: I don’t think I’m anywhere near motris level on these as my observational puzzling skills are a little low, but I always do fine on them in practicing USPCs. I just redid last year’s Takeout, which had the same rules, and figured that was enough.

On the test: One of the most well-crafted word searches I’ve seen. It was pretty neat how so many words crossed their own missing letter spot, and it made things trickier since you couldn’t just find a word, get a new letter, and use it to get the next one.

I should note that during the actual 2010 USPC, I was specifically looking for a word to appear in the set of blanks. I kept that habit here, especially since the constructor was being so tricky in other ways, so STARDUST didn’t stump me for long.

Overall, I felt like I was a little slow on this one; there was a lot of staring, especially at the end with a couple letters to go when I couldn’t find ROOSTS.

Time: 9:43 (0:42:31)

13. CORRAL

Preparation: None. I’ve always had a fairly easy time with them in past USPCs, and I was guessing this would still be a nikoli one instead of a Dave Tuller design. nikoli’s have a better logic flow, which fits my solving style very well.

On the test: This one is the easiest Corral I can remember appearing on a USPC; those 2s made it really easy to start. It was interesting how row 7 was almost all outside. I’m used to macro steps based on not sealing a set of outside-the-loop squares in, but I don’t see avoiding separating two large interior regions as much.

As has been noted many times, the answer entry for this puzzle is particularly mean, so almost a minute of my time was spent verifying it. First I entered it, then I counted up all the outside squares by row and verified I got the sum of my answer entry, then I checked my entry for typos again.

Time: 2:40 (0:45:11)

14. BLOCKS

Now we’re getting to the really meaty part of the test. This part is a bit more heavy on the logic puzzles, so it’s where my performance tends to shine more.

Preparation: This one had me seriously worried. A lot of this owes to the practice puzzle on the AOPS blog post, which was a nightmare that took me a long time. Quick recommendation to anyone who tries it: there is a surprising amount of logic in that puzzle, and the more intuition you try to use the harder you’ll find it (that was my problem at least).

On the test: This one having many more walls than the practice one above was hugely encouraging. It was pretty obvious where to put the square block. After that, the bottom left configuration of walls caught my eye, particularly because it had exactly two exits and could not be fully traversed if there was no piece there. I eventually reasoned that a line piece in the corner was the only way it could ever work. Then I saw the dead-end in the lower right, tested all the ways to fill it, and got only one that made much sense.

At this point there was a lot of the loop I could draw in pretty certainly, and in the process all the other piece locations just gave themselves away. It might have helped that Simple Loop, a type where you just draw a loop through all squares, is one I’m really quick at. Another confidence-boosting solve.

Time: 2:30 (0:47:41)

15. HEX WORDS

Preparation: For this puzzle, Word Connections, and Jumping Crosswords, I was kind of reminded of the kinds of word puzzles that appeared on past OAPCs. I went through all the test instructions in the archives and printed the types that looked the most similar. They were Island of Numbers (OAPC3), +/-1 (OAPC7), and Tetroword (OAPC9). Word Connections and Jumping Crosswords turned out to be Cihan Altay, but this puzzle was Serkan’s, one of the main OAPC constructors, so when I saw his name I figured I had made a pretty good choice in practice material.

On the test: I felt a little less confident when I had so much trouble working into this one though. I noticed that the Z and U had few options, and that UPDATED didn’t fit the U, but I couldn’t get any placements. After a couple minutes I put it in the todo pile.

16. WORD CONNECTION

Preparation: See 15. In particular, I figured Island of Numbers would be particularly helpful for this one; it would just take Numberlink skill instead of Nurikabe to fill things in. And I have no shortage of confidence in my Numberlink.

On the test: I spent a bit less than a minute toying around with what that P in the top right could be, settling on SEPTEMBER. But I felt like the guess space was way too big, so I erased it and tried to look for more logic. I noticed none of the grid’s givens were such that one month could use two of them. So there’s going to be 12 of them! Oh, wait, just 11. Perhaps I can determine which one is unused. Yes, it’s JUNE!

Okay, how about trying to figure which letter goes to which month? MAY had to be the Y, then JULY had to be the L… I continued going through the months in increasing order of length. It turns out you can figure out (up to duplicated letters) where every month goes as a first-level deduction.

Then I turned on my Numberlink sense and started drawing in the months in order and paths. I had some bad intuition at first and ran into trouble at two points, but in every case I got out quickly by tweaking the JANUARY-FEBRUARY path, which was quite a tricky one. This solve felt relatively quick, due to my strong Numberlink sense.

A lot of checking went into this one’s answer entry. I used parity orientation of entrances/exits of adjacent months) and even counted up all the turns in the puzzle by row. Luckily I didn’t need any corrections.

Time: 11:20 (0:59:01), which includes some time spent staring at Hex Words and at least a minute of checking

17. HOPPER

Preparation: Just motris’s preview puzzle. And boy was that a confidence killer. I never actually solved it, or had any idea of what this strategy he mentioned was – after the test he said it was about hiding 1s and 6s, which I still don’t get. I had this one marked, along with Dynasty Sudoku, as the two things to go for only when almost everything else was done for fear of dropping lots of minutes to them. Those are the two types motris previewed; I’ll leave whether this is a coincidence or not to your imagination.

On the test: Expecting to put this one off quickly, I was highly encouraged by having half of the pieces as well as bigger ones. I spent some time looking for some ways to match some of them up, but after an idle minute sent it to the todo pile anyway.

18. CORRAL CRATES

Preparation: No red numbers, but geez, this is my type! I figured since my type was similar to a Zotmeister variation that this would be one of his puzzles. He always has nice logical paths, another reason not to be concerned. I did my all-black Crate Corral for practice and called that enough.

On the test: Oddly enough, it was constructing another blog puzzle, the large square Sunday ordinary Corral, that gave me an edge here. Not because of the size. The use of low numbers (like 2) diagonally adjacent to other clues was in full force here, and I had explored that thoroughly in my previously mentioned puzzle. Thanks to that, getting in took no time at all, and I never ran into any trouble later. Usually constructing puzzles doesn’t help so directly like this and the benefits show in other ways, but sometimes it can. Another confidence-boosting quick solve.

Just like on 13, I spent about a minute verifying my answer here. I had to be careful not to count crates when counting outside squares by row though.

Time: 6:00 (1:05:01), which includes a bit of time spent staring at Hopper

19. KAKU ROGUE

Preparation: I wrote a casual puzzle test at a summer program I worked at this year, and it had a sign-based variant of Kakuro, although its rules weren’t that close to this one. Actually I wrote two of those puzzles since the first one was too hard. So I redid both of those. Then I did the signed one on an earlier USPC. Finally, I started constructing my own 9 by 9 (the answer entry info in the instructions gave the size away!), but didn’t get very far before deciding I probably wasn’t going to find any more logic than I already had.

On the test: This one took some thinking, but you can count on a Michael Rios Kakuro variation to always have a simple logical step waiting to be found when you get stuck. With that in mind, I was able to keep myself from losing the trail for too long. The sectioned off corners, which allow for the logic of “add rows, add columns, take difference” step, made for some easy break-ins.

It also turned out that my practice puzzles had in fact helped a lot. The puzzles I written had a lot of the patterns about getting extreme sums with negatives showing up. For instance, if you have a low sum clue and know it has a small negative number in it, there are a surprising amount of constraints on what that number can be. A 3-term sum with a -1 always evaluates to at least 4 for example, and with a -2 it always evaluates to at least 2. My Kakuro is not great, especially compared to motris, so I didn’t feel awesome about this solve, but I was very happy to have it done in a reasonable time.

Time: 6:52 (1:11:53)

20. JUMPING CROSSWORD

Preparation: See the note under 15; +/-1 was particularly helpful here. Later I saw motris post some difficult reference material from 24HPCs. I did both of them, and they were incredibly good practice. I felt really good about my chances here after having churned those monsters out, and I had a good idea of what to look for both at the beginning and in the middle.

On the test: The most important things to keep track of for this puzzle are the word lengths which are long but have a short bank, and first/last letters, roughly in that order. The first heuristic wasn’t so encouraging here; the best was the ones of length 8. I found where they were in the grid, but couldn’t see much to do with that (if you read motris’s report he seemed to have though).

Also, I read on a forum that someone was encouraged by the long word lengths and tried this puzzle. Actually, the word lengths here are anything but encouraging. The two lengths you’d want to see are either no blanks at all (even one is terrible), or less than half the length so it’s blank-letter-blank-letter…. And even in those two cases, I don’t think there’s much you can do in terms of break-in material.

After some more staring, I eventually put this one into the todo pile.

21. HUNGARIAN TAPA

Here we are at the really high-valued ones at the end.

Preparation: My tapa solving is quite good. Unfortunately back when the TVCs were held at LMI I hadn’t done much to try to cleanup my tendencies for errors, and I was breaking a lot of the puzzles (sometimes noticing, sometimes not), so my results were pretty poor compared to what I thought I was capable of. Here was a chance for revenge. Having that confidence, I stuck with just redoing the three on the TVCs.

A lot of doing this type is just keeping standard Tapa tricks in mind and remembering how to form sums in Kenken, with repeated numbers allowed if they are in a different row and column. One slightly less obvious piece of logic is the fact that you know exactly how many shaded cells are in each row. Be on the lookout for rows/columns that have almost all of their black squares or white squares, and use them appropriately. Sometimes for almost-saturated rows/columns you can bring in the two by two square rule to nail down where the black cells have to be.

On the test: I figured with the worth of this one, it would be a bit bigger and harder than the ones I practiced, and I was right. There were a lot of placements to get at the top, although not too many actual numbers. After bumming around there for a bit, I decided to move down to the bottom, where I got some more things.

Eventually the 6-26 caught my eye and I realized that was as high as it could go, needing a 6,66554 around it. Then it was just like a tapa 1-5 clue, with two possibilities. I tried one, got a contradiction, tried the other, and… got a contradiction. After holding back panic, I eventually found that I had used some bad uniqueness logic in the first case, and ended up having both contradictions combine to give me tons of progress. While I’m sure there are cleaner logical ways to do this puzzle, I was confident that I was going to get something out of this bifurcation, so I went for it right away.

There was nothing too hard afterwards, but it definitely took some time to finish. I was also sure to check my solution due to the high point value and time spent.

Time: 14:07 (1:26:00), which includes some time spent staring at Jumping Crossword

22. DYNASTY SUDOKU

Preparation: I started constructing one of my own around the time motris’s poll wrapped up. It probably would have been an 80-pointer or something with all the ridiculous jigsaw Sudoku steps I was stuffing in there. Later on I found a global contradiction where there was no way to place all instances of a certain digit, and things had probably been broken for awhile. I couldn’t tweak it at all, so I decided I had learned enough from the progress I had made and scrapped it.

By then motris had posted his, so I did that one. As he’s said in comments, there’s a point in the solve where you shift to finding black squares, and that indeed held me up the most. I also had some bad logic near the end and spent a lot of time tweaking. So in the end it was a really slow solve.

Thinking about this puzzle was my biggest worry pre-test by far. I was 99% sure this was a Zotmeister puzzle from how he always has a Sudoku variation and his love of the dynasty style (the fact that he’s one of few to use that term also helped). Memories of his Sukazu and SuDUOku stumping me the last couple of years came to mind. Furthermore, I have had little luck with 40 point puzzles in practice runs, particularly Inside/Outside Corral of 2009 and Ampersand Criss-Cross in 2008. Tons of time would go into them, and no points would come out. And yet… this puzzle is 40 points. Can I really just throw that much away?

I eventually decided that I would give it my best shot on the test, but be ready to move on if I ever felt like it wasn’t going well.

On the test: Yes, it was a Zotmeister. Starting out, I managed to get a lot of placements pretty quickly. I think my main work-in was the two dumbbell shaped rooms. In the case of the top one, its bottom two cells had to match R3C2 and R3C10 setwise by some Jigsaw Sudoku logic, and similarly on the bottom. Also, R3C1 and R3C11 match setwise with the top cells of the top dumbbell, and similarly on the bottom. That was a lot of placements, and it also got me some black squares in the dumbbells. Progress from there wasn’t quick, but I was never getting stuck enough to get even close to putting it away.

As the constructor himself noted in a comment on motris’s blog, this probably wasn’t hard enough to be worth 40 points, so I’m very glad I didn’t put it on the shelf. Also, I benefited heavily by the fact that it used a lot of the dynasty logic, which I am much better with than Sudoku. This was not the case with my failed puzzle or motris’s.

After a lot of tricky dynasty logic and several subtle pairs later (but no guessing), it was done. Due to all of my worries pre-test, I was pretty ecstatic about this. I gave this one a very careful check before moving on, and I scanned every row and column. Can’t drop 40 points, especially when that much time was spent.

Time: 10:25 (1:36:25)

END OF FIRST PASS

I’m at a maximum possible score of 300 points, already higher than my 285 last year (which fell to 225 since three errors sneaked in). Quite encouraging since I had a little less than an hour to go. My todo pile has just 5 puzzles, way lower than I usually end up with after a first pass. This too was encouraging; I usually have a number of partially solved puzzles from the latter half to deal with, but now I was finishing everything I was starting. I just had 4 (differences), 8 (counting), 15 (hex words), 17 (hopper), and 20 (jumping crosswords) left to do. I was sure I could get 15 and 20 out, but the others had me a little anxious.

4. BARN STORM

Take 2: I caught a whole bunch of differences near the upper left corner, with A1 being almost immediate. I ended up with 5 in that area, but no matter how much I looked in the other regions I couldn’t see anymore. I entered those 5 (but evidently didn’t hit submit) and put this one away again.

8. SQUARE COUNT

Take 2: I first did a detailed check of all the stranger lines, neither straight nor at a 45 degree diagonal. After convincing myself there were no squares there I started counting them by size and orientation. 9, 4, 1 straight squares, just as in the example.

I turn the paper 45 degrees, and the diagonal squares all became apparent. 14 of the small ones… now for the 2 by 2 where the center screws it up. I noted that it was a full set of lines except that one segment in the center. Somehow in the process of noticing this I caught wind of something funny near the center segment… it was a tiny white square! Very glad I caught that one. Almost put down 2 of the 2 by 2 squares, but eventually remembered that all of the lines in one direction were there and got 4. It was easy to see there were 2 of the last size.

At this point I was kind of worried. That went by awfully fast, but I couldn’t think of any that I had missed. Eventually I went ahead and totaled things up and put down 35.

Time: 5:28 (1:41:53), about 3 minutes for half of Barn Storm and 2 minutes for Square Count

17. HOPPER

Take 2: I thought about the strangeness of the piece with the 7, what with there being a strange branch in the possible paths at the bottom. Some more reflection later made me realize the whole piece worked nicely if I could nest a 6 in there. Sure enough, one other piece had a lonely 6 that was isolated from the other numbers on the piece, so I put those two together on a piece of graph paper. From there, I fairly smoothly found ways to keep the paths going with new pieces, and was shocked to have a full set of working paths after just a few minutes.

I hurry to fit it into the grid and… uh-oh, that top row’s not going to get used… aw crap, I need 9 columns! Yet it felt so right… this has to be a tweak away! A few minutes later, I didn’t find it. Rather rattled, I decided to move on and see what I could do later.

15. HEX WORDS

Take 2: A couple more minutes of staring didn’t yield any logic. Remembering how quickly these criss-crosses fall if you got started on the right foot, I decided it was time to throw down a guess, and printed out a second copy in advance. I went with putting the longest word (UPDATED) somewhere, and figured since P was not a common letter I should go ahead and use it. STELE worked wonderfully with the two Es near each other, and then BRITT used the two consecutive Ts. Definitely right; boy did I feel lucky. The rest was pretty smooth, although not lightning quick. Again, as I mentioned earlier, doing a full volume of nikoli Nansuke is a good way to learn how to keep making progress on these – at least it worked for me.

The statistics show this was the least-solved puzzle on the test. I suppose I can understand that. It’s a challenging variant and the puzzle itself has no easy way in. It also seemed a little undervalued for how long and difficult it was.

Time: 14:05 (1:55:58), includes several minutes of almost solving Hopper

17. HOPPER

Take 3: Back to my 9 column 7 row almost-solution. I briefly thought of starting over, but I seriously didn’t think I could get anything too different from what I already had. How could that 6 not go with the end-piece, for example? So I settled for trying to tweak some more. And eventually it did tweak! The makeup of the path changed significantly from what I had thought it was, which is probably what stymied me before, but I managed to move a couple pieces to make it fit just right.

Thanks to my earlier debacle of writing the earlier wrong answer onto the page, I ended up with a written solution that went off the board a column and left the last column blank, which made answer entry a little tricky. You can bet I checked that one thoroughly.

Time: 2:51 (1:58:49), includes just time to tweak

20. JUMPING CROSSWORD

Take 2: I mentioned above that the two things to go for are long lengths but short banks, and also first/last letters. The first one didn’t get me much in my last try, so even though I never used it in my practice puzzles as a break-in, I thought I’d try the second heuristic. From this I quickly caught what I was sure was the key to the puzzle: those 12 length words all ended in one of three letters.

There were some 4 and 7 length blocks at the end of the 12 blocks, and I was able to narrow down their possibilities pretty significantly, getting that the two 4 blocks had to be ONE and END between them. And then that O got me in: only one word in the whole bank ended in O. That pinpointed which block was ONE and which was END, as well as the location of DISCO, the word that ended in O. Thanks to the 24HPC practice puzzles, progress was quite steady from there.

Other people got into this one through the top left corner, where apparently the word going across could be determined at the start, or as mentioned by using the intersections of the eight letter words.

Time: 12:41 (2:11:30)

4. BARN STORM

Take 3: Frustration hit a maximum here. In doing tests for practice there’s been at least one occasion where I’d have just Differences left and 40 minutes or so, and I couldn’t finish. I was going to be pretty upset if this one stopped me. Yet, one by one, those five remaining differences fell, though each one felt like it was taking forever.

My last one was the different lengths of lines in A3, in that top left corner I figured I had cleaned out in my first batch of differences. That A3 one got me because in past USPCs I’m used to getting 9 differences on these and then having the last one be a length thing that I just can’t find. It seems like there’s always exactly one. Here, I had already found the one in the top right corner and figured that must have been it. But there were two length changes this time! So that last one took me quite awhile.

Fingers were trembling as I entered those coordinates…

Time: 10:05 (2:21:35) for last five differences

FINALE

That left eight minutes of checking. After each puzzle I had done a pretty thorough job, counting total turns or total outside squares in the whole grid for Word Connections and Corrals for instance. In a USPC practice run when I had done that much I ended up free of errors. There was just one puzzle that needed a thorough check: Square Count, of course.

First I checked my subtotals. And what a good idea that was! There were 16 small diagonal squares, not 14! It wasn’t that I missed seeing any, it was an inability to count to 8 properly (I was using the line of symmetry). Just 33 seconds after finishing Barn Storm, with 2:22:08 on the clock, I put in 37. Then I thoroughly redo the puzzle, and get nothing different. That left about enough time to give every puzzle a quick glance and make sure I didn’t typo on the answers. I found no changes to make. I got through checking the last one (Jumping Crosswords) when time had already run out, and so that was it.

I never broke and restarted a puzzle, I had several lucky guesses (especially 10), I never got too stuck trying to figure out why I’m getting a contradiction and no solution, which is a frequent issue of mine… so I figured I had done well, and that it was a performance I’d have trouble ever topping in the future. (This might explain point 3 of my facebook/google+ status that some of you saw.) I was sure I was going to see motris post that he finished 10+ minutes early or something like that, especially since I had lost such an enormous amount of time on the Differences, and since I couldn’t touch him in USPC practice runs for 2007 and beyond.

That didn’t end up happening though. It’s impossible to say whether he would have done that well if he had been as good of form as he has been in past years, but I think it’s pretty clear he underperformed a bit. So I definitely feel like I have things left to prove, and I’ll be eager to try to defend my ranking from him next year. I don’t believe that another win will come easy.

As always, thanks to Nick Baxter and everyone else behind the USPC. This year’s test was outstanding, with rules changes that greatly benefited the contest and a level of puzzle quality that stands on or above the level of the great tests LMI has been holding. I look forward to trying to bring this level of performance to this year’s WPC and to defending the win on next year’s test.

Advertisements

13 Responses to “Detailed USPC Post-Mortem”

  1. Zotmeister Says:

    “this was a Zotmeister puzzle from how he always has a Sudoku variation and his love of the dynasty style (the fact that he’s one of few to use that term also helped)”

    True story: when I submitted Dynasty Sudoku under that title, Nick asked me about it, as he hadn’t heard the term ‘dynasty’ for “Japanese-crossword rules” before, but liked it and wondered how established/accepted it was. I explained how I invented it and that I was, to my knowledge, the only constructor to use the term, but that no one else has suggested an alternative during the six years it’s been out there. Nick decided he’d go with it and see if it sticks. Checking the blogosphere post-test, it clearly has stuck. – ZM

  2. Thomas Snyder Says:

    My hopper description may have confused because my use of language is not always precise but a lot of what you said above suggests you figured it out during the test if not before. “Sure enough, one other piece had a lonely 6 that was isolated from the other numbers on the piece, ” Those cases are what I meant, and perhaps instead of the word hiding I mean get the 1’s and 6’s connected somewhere on the grid so that the whole path works near it. Because there is likely just one 0 and one 7, almost all 1’s basically need to abut two 2’s and all 6’s basically need to abut two 5’s so those are the easiest numbers to apply counts to and try to connect to things.

    The first step of my practice puzzle is the “obvious” form of this where an isolated 0 needs a lower-right 1 and there is only one such piece. This actually strands a 5 so you need at least one piece that has 6’s or 4’s and that P with 2 6’s really needs connections. The greedy approach here with a perfectly shaped upper-right corner will put those three pieces in but I can describe the logic to skip that greedy approach. Other problem spots involve
    getting two different 5’s near the lonely 6 on the V (it cannot be one piece). The 1 on the I and the 1 on the Y also need to connect to something and after many placements you are somewhat 2-poor.

    • MellowMelon Says:

      Ah I see. In the case of your puzzle I had found the step with the U almost immediately, and that was basically as far as I had gotten for certain. I think the next thing I did (early in the solve) was put the P in exactly as you suggested, but when things went wrong later I eventually took it out and even found other shapes that could fit okay in there.

      In fact the whole time my approach was similar to yours; identify space that feel stranded and use it to find excellent fits between pieces. The search space on that puzzle was just so big (maybe more so than you realized) that I never got the ones the solution needed. I should also mention that damn I pentomino, which had the high numbers on the right instead of the left, was ticking me off the whole way. I think I eventually caught that it worked well next to the Y, but even that I think got reconsidered later after some failed guessing.

      • Thomas Snyder Says:

        Oh, I realized the search space was big (although it could be even bigger than when I was test-solving it). But I’m also used to USPC puzzles that are uncomfortably computer generated with large search spaces (Dot triangles anyone?). So I made something pretty hard, where actually packing the space was a much bigger meta-requirement than anything we ran into in the actual puzzle. It being from Serkan and not Erich certainly helped.

      • Thomas Snyder Says:

        It also never became relevant on the USPC version, but if packing the space is a big constraint, you want to view the whole space like a checkerboard with even/odd tiling. There are many puzzle types where parity is the big thing and my prediction was that this might be the case here as it is inherent in the design. Certainly if you wanted to pack corners in my puzzle, you could find that even top corners are pretty likely (forcing odd bottom corners, although neither ended up being used). But odd top corners and even bottoms is much less likely. Even if a shape packs nicely, it must obey parity shading so only odds touch evens and vice versa.

  3. Bram Says:

    Always like reading about how people solved puzzles and compare how I did them.
    I find it funny to read about how everyone solved the jumping crosswords. I skipped it in the test, but did it afterwards as i missed the opening to it at first and went for the dynasty sudoku in the last 15 minutes. I started with the word FORM, as it’s a full fill and it had a rarish letter. It’s easy to find its place and it gives a lot of information as you get 4 starting letters.

    • MellowMelon Says:

      I think the fact that you missed said break-in during the test may be a confirmation of my note about the role of heuristics. But still, nice.

      • Bram Says:

        As a non-English native, heuristics is one of those words I’m never really sure what people mean by it, Just like when people use the word metalogic. I understand the meaning on certain occasions but usually not really sure what it implies.

      • MellowMelon Says:

        Ah, sorry. By heuristics I’m referring to the ways in which one looks for and finds a solution. For example if someone gets stuck in a Sudoku they might decide to look for naked singles, and specifically they might focus on rows/columns/boxes with four or more numbers already specified. Someone else might look for something different. Both might end up finding the same naked single to keep going forward, but what they did to arrive at it may differ, and it will affect how fast each of them solves the puzzle. This is what I mean by heuristics; it is fairly different from the steps by which someone solves the puzzle, if you’re keeping solving time in consideration.

      • Bram Says:

        Okay. I find with logic word puzzles, there’s hardly ever a single path towards the solution.

        For this puzzle, I just never really got back to it. It looked too big on first glance and I had printed the test back to front by accident(new printer). So on my first printer check I had the last 7 puzzles and skipped it quickly as I figured I could faster do some of the easier ones first. Eventually only solved the Dynasty Sudoku from the skipped puzzles.

  4. TheSubro Says:

    Thanks for the insights into “A Beautiful Mind”!

    In very few contexts do people greatly interested in a “hobby” get to dialogue with those who are the best at it, and gain tips about how they do it and how to improve. It makes being party of the puzzle community that much more enjoyable. Thanks for taking the time to share.

    TheSubro

    p.s. Was traveling on Saturday, so I could not compete this year, but did it off-line under “game-time conditions” the next day. I enjoyed it greatly, and was happy with my performance. Many thanks to you, Motris, LMI, Mathgrant, detuned radio, Nikoli and the like for all of the fun extra practice afforded of late, and for making us all better..

  5. Rob Says:

    Congratulations!

    It’s good to see other people were having trouble with Hopper. I’ve not yet solved either motris’ or the USPC puzzles.

  6. Serkan Yurekli Says:

    First of all Congratulations Palmer for title and amazing performance.

    I always try to place a logic solving path into my puzzles. Actually it is more fun for me, I think for solvers too.

    Amount of Z and U is starting point of Hex, next step is Judo placement etc. When I prepare Hopper I wanted that solvers should consider grid size to solve puzzle. As you will notice, if this puzzle has only one solution, then the constructed shape (combination of all pieces) cannot move to any directions. So it means that each 4 sides of the construction must touch the edges of the grid. After processing with pieces, if solver figures out this, he/she could solve easier. This was one of my purposes in this puzzle. I am not sure whether it worked or not.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: