19 January 2018

This is a followup to my earlier post recapping this year's Mystery Hunt. I want to go through each puzzle that I really worked on during the hunt and give my experience with each of them. This post will be heavy on spoilers so stay away if you're thinking of solving some for yourself. The puzzles here are ordered approximately by chronological order in which I worked on them.

## Yeah, but It Didn't Work!

My part on this puzzle was pretty straightforward. We identified some lettes and got the phrase. I'm not sure who connected it with the Have You Tried document (It wasn't me), but that was pretty cool when I heard about it. I really like the way that the first few puzzles in the hunt doubled as pointers to good resources for newer solvers.

## A Learning Path

Continuing on with the introductory puzzles, I thought this was pretty nice. I did a count of the letters in each of the paragraphs although I was rather confident going in that they would match with the grid sizes. Someone else made the Hamiltonian path connection, so then we mainly just had to solve the logic puzzles (which were actually pretty fun, despite how simple the rules are!). I think I did one out of the six and did the final extraction as well.

As a side note, the one puzzle I did coincidentally gave the same yellow square order reading horizontally as reading along the path, so I was a bit confused about which order the puzzle was using, although the answer that came out was obvious enough.

## Just Keep Swiping

This puzzle was unfortunate. We noticed the "add L/R" mechanic and identified all the screen names, and drew out the tree. Then we indexed into the tree by the path represented by capital letters on the leaves, but then we were just stuck. After a little while I moved to other puzzles, and I don't know if we eventually forward solved it or backsolved it.

I'm generally not a fan of indexing by level of a tree. It feels very artifical to me because there's usually (as is the case in this puzzle) highlighting that the level number is important. This isn't the first puzzle to use that extraction, and so it's a blind spot that I'd like to fix for my personal solving skills, but it's something I would try to avoid as an author. I feel like the clue phrase should have just been spelled out in first letters once you do the indexing into the tree. The main worry is that it might allow skipping the index-into-tree step, but I doubt it would cause a problem in practice (If it was tried that way in testsolving and too many people skipped it, that's a good reason to have the puzzle this way!).

## Games Club

This gave me a nice throwback since the metameta for the 2009 Mathcamp Puzzle Hunt included the phrase "USE CHOMP NIMBERS". Sadly the extraction here was number of turns instead of nimbers, and the chomp game is rather trivial because it doesn't include the poison square, but it was a fun exercise. I wasn't sure how hard it would be to solve the 4x4 domineering game, but I was able to look it up and use the context of the puzzle to figure out how many turns it must last. It's interesting to me that the solution calls out Wythoff's Game and Chomp by name, but not Domineering because the latter has much more intricate theory.

## Anger Meta

Our solving experience for Anger was amusing. We pretty much knew exactly how it would work from the beginning, but the distribution of temperatures required us to unlock more puzzles to be able to know which columns to put answers into. It was convenient that the temperatures had equal sized gaps, so we were able to know where the last puzzle would go in the grid despite not having it unlocked. But somehow we hadn't solved it. That's where I stepped in, and noticed that we should use RAGE instead of ANGER. A short while later, it was done.

Overall, the emotions round was mostly uneventful for experienced solvers, but the puzzles were still clean and well executed. We finally got our first island unlock, and decided on Dark, which turned out to be the Sci-fi round.

## Fuch-sia

This was pretty straightforward. The logic puzzle part at the beginning was exactly what we expected, although it was amusing to get lyrics out like "dirtypurple eyed girl". Then some people tried to translate color to hex codes via Google, which of course gives results but nothing canonical. Apparently the xkcd color post is a lot less well-known than I thought, but once I pointed out that that would be the correct source the puzzle fell quickly.

As a member of the we-tell-you-the-rules logic puzzle genre, this puzzle doesn't have a lot to say about it. The puzzle type was quite interesting, and I'm always impressed when logic puzzles like this are written well. A lot of people got stuck on these puzzles, so I think I personally solved two of them.

Because we were getting stuck so much, I decided to look at extraction when we had (I believe) four of the seven logic puzzles, giving us four bigrams. Of course, the answer turned out to be two words, so a normal word list won't contain it. However, I had spent a good part of the week before hunt building a bigram dictionary based on Wikipedia articles, and the answer showed up enough times to be included. Preparation paid off! This was the only puzzle where the bigram dictionary ended up really helping me out, but I was happy it happened even once.

## Self-Referential Mania

This puzzle is exactly what you'd expect. My role here was pretty minor, as the bulk of the logic was done before I joined. I caught a couple errors, and also planted the (wrong) idea of using Spanish scrabble values for the second half in addition to the Spanish number names. But I was also the one to cut that idea down, and noticed that "quince" would work for everything we needed if we used English values.

## Flags o1 6ur 10th15

Other people translated the flags into letters, and also identified a bunch of the categories, but we had to figure out what the heck these numbers meant. Obviously the title was supposed to somehow correspond to "Flags of our fathers", but it was unclear exactly how. Eventually I said something like "F is the first letter, and O is the sixth letter," but I didn't really believe it until a teammate convinced me that that was the actual mechanic. Then I was the one tasked with writing the code to evaluate all of the expressions, and we had a bit of confusion with whether Lyon Sprague de Camp's last name is "Sprague de Camp" or "de Camp", but eventually we figured it out and were able to extract in the obvious way.

## AllSpark

I don't know exactly how fair it is to say that I worked on this. A teammate and I started looking at it with two of the feeder answers, and he figured out the letter transformation mechanics. I helped with grepping and we found PULSAR and ROGUE, but we then had to leave the puzzle for lack of more answers. Then later other people did the rest. I hope our work was at least somewhat useful!

At some point while we were working on Sci-fi, we also unlocked the Pokemon round. As some of my readers will know, I have played a lot of Pokemon, and held the speedrun record for Red/Blue Glitchless for a while back in 2013. So naturally I was excited to work on this round, and I basically ignored all the other puzzles until we were done. As a result, I didn't do anything in the Games round at all!

## Shoal Patrol

Once again, a we-tell-you-the-rules logic puzzle puzzle. I think I got to solve two of the grids. CHAMPION missing M stood out to us a lot and I helped find some of the others via anagramming. I don't believe we ever made the connection to Avenger class ships. In any case, I think we were missing one or two at the time of extraction but YARDARM was a good enough answer to call in.

## Submarine Patrol

So then we got this puzzle, and it was glorious. Obviously there were no real tricks in how it got solved. I decided to work on the battleship puzzle independently from two of my teammates, and got most of the way but they finished first and it was clearly inefficient for me to try to add a third head into the minesweeper and path parts, so I went to other puzzles. Still, the evolution of Shoal Patrol into Submarine Patrol made me very excited to see the rest of the round.

## 33 RPM

I see an image of a record, with a very clear spiral oscillating between black and white, so of course I should be getting an audio file out of that. So I wrote code for that and posted it in the slack channel, and mostly left since I'm completely useless for identifying songs. We got to "9 times 43 times 64 over 49" pretty quickly, but couldn't figure out what to do with that for a while. We called in a bunch of things related to 505 until someone finally had the idea to play the file at 505 RPM, and the answer spilled out.

## Twitch Plays Mystery Hunt

So like I said, I had left 33 RPM, and when this puzzle opened up a huge number of us were immediately drawn to it. We had some fun just messing around for a little bit, but then quickly decided to organize, and I somehow ended up as the commander, so I got to yell at a group of about 10 people what to do. Unfortunately, when we were about halfway through the first level, some other people were looking at Mix Tape and needed me to do the image to audio conversion...

## Mix Tape

Okay, so it was reasonably straightforward to modify my program from 33RPM, although I had to mess with the playback speed a lot to get something recognizable as songs. Reading the solution, I'm disappointed that the authors intention was for us to reuse the 33RPM from the previous puzzle. Unlike record players, cassette tape standards are in inches-per-second, so time should correspond to arc length rather than angle. Since I did that while the authors did not, our audio was time dilated based on the distance from the center of the spools.

In any case, once I had produced the two audio files, I left again because I'm still useless for identifying songs. So other people got the phrase "Mirror and take away right half from left," and then got stuck. I did a bunch of audio manipulations expecting another song to come out like in 33RPM. When nothing was sounding good, I decided that it was going to be less error prone to do the manipulation to the image and then translate it into audio. Of course, after doing the image manipulation, the answer just popped out.

Overall, I really liked this pair of puzzles, but I wish a couple of the details in Mix Tape were slightly different.

## Under Control

I didn't do much on this puzzle, but I wanted to highlight it because I think it didn't go the way the authors intended. We did the ninja stuff and (I think) had a good time, but then we had the names of moves and got stuck on extracting from it. Eventually this got backsolved because we noticed that "RIGHTMOST" contained the letters of "GHOST", and tried removing them (first to get "TRIM", reading as "riGHtmOST").

So this was definitely a case of getting stuck on an unfun part of the puzzle. I think the main culprit is the mechanic of diagonalizing on a word basis. I don't think I've ever seen a puzzle where that diagonalization mechanic is truly motivated. Add in the fact that the RPS-cycle-order is about as natural as final-battle-order (especially since teams didn't have confidence that the final battle order was set), the extraction was particularly hairy.

## It's Not Easy

This was another puzzle that ended up backsolved. We did the crossword and started by noticing that the rows were in the "wrong" order, and then once we filled the whole grid realized that it was just bottom to top. Then we realized that the Green building had the appropriate number of windows, so the grid should map to those windows. And then...nothing. Our first scout brought back a picture and none of us really saw anything special. A few hours later I went out and confirmed that the set of lights that were on were different, so I expected that there wasn't really anything to get from the lights. Upon review after the hunt, the supposed arrow window is extraordinarily blue in my photo, but I couldn't really see an arrow. I eventually backsolved this while we were stuck on the final round in Hacking and just wanted a change of pace, which was well after we solved the Pokemon metameta.

## X Marks The Spot

Another puzzle where I didn't contribute a whole lot. I heard people around me talking about triangle centers, so I wanted to see what puzzle they were talking about. I got "ISOGONAL" from grepping when we had maybe three of the letters, but of course since it wasn't directly the answer we had to continue identifying the centers, and then eventually build up enough confidence to determine that ISOGONAL was a clue to take the isogonal conjugate of all the centers. It was fun, though!

## Pokemon Evolution

So the Pokemon Island meta-structure was my favorite "puzzle" in this hunt, although I don't know how much it really counts as a puzzle. The first evolution mechanic we got was the Fire type, as one of my teammates noticed that all of the answers appear in We Didn't Start the Fire. The Rock mechanic was pretty clear from Googling the first two answers together, and someone else noticed the Drag-Dragon connection. It was also one of the more fun ways to do backsolving: "What's perpendicular to a wall?"

## The Taxonomist

Before we actually get to the metameta, I should talk about the two submetas that I helped with. The Taxonomist was the first mechanic I expected, although I almost thought that "National Taxonomist Dexter" was referring to something else and then realized that it was just a play on "National Dex". One of the interesting features of this meta was that the operation chain -> ^3 -> /7 -> x6 -> had to start with Squirtle, since 14^3 is too large. So we had backsolving information for It's Not Easy really early on, and we still got stuck.

## The Scouts

I want to make a note on that extraction, though. I think that there's a wide range of how people feel about what I'll call "random diagonalization". When I started doing puzzles through the Mathcamp puzzle hunts, I thought that diagonals were something you should just check in general. After getting to MIT, some of the people I met there were more of the opinion that diagonals are only reasonable if the letters form a square. Clearly this writing team is quite happy to use diagonals in non-square contexts. I'm not particularly opposed, but it's something to keep in mind for future writers, and I think that some of the solvers' experiences in this hunt (especially on Under Control) really demonstrate the potential downside.

So now we get to the actual metameta. The Sotomayor/Vinson Justice connection was the first one we noticed, but we didn't know what to do with that for a while. We floated the idea about trying to reverse the evolution, since each of the Rival's Pokemon was an evolved form (although Flareon evolves from a normal type), but eventually noticed that ONE contains a letter from GHOST so it can't possibly work.

The first answer mechanic we got was from Black Magic, where we figured out that the Muses were a cycle in Legions, and got Graveborn as the black one. That also gave us easy backsolving, as the two cycles in Stronghold were Licids and Walls. From there it was a pretty short leap to figure out that Be Mean refers to averaging, and Take Out refers to subtraction.

And now, since like I mentioned I didn't do anything in Games Island, on to Hacking. Like I mentioned in the last post, Hacking was our last island, and with the narrow unlock structure we were mostly cramming our whole team into three puzzles at a time until the full unlock, which was less than pleasant.

## The 10,000 Puzzle Tesseract

As an obvious throwback to The 10,000 Puzzle Pyramid from our hunt, this one caught my eye right away. Probably the biggest difference is that the files in this one were much easier to parse than in the 2015 puzzle, which allowed either exact numbers or ranges, either absolute values or percentages, words that got an s when pluralized, and so on.

Of the functions, scrabble and units were obvious, and typewriter, elements, and news didn't take too long. We were stuck on midpoint and index for a long time. One of our team members noticed that midpoint should be something similar to center of mass, and then I figured out how to make it work correctly (by not weighting letters by distance from the pivot). I don't remember who figured out index, but I helped a little bit with figuring out how to calculate it quickly.

So then we had a list of 10,000 words. I think I went to sleep around this time, and when I woke up I was told that we weren't much farther, except that we had noticed that the clue order spelled out "UNISTEMS", which apparently means anagram +1 letter in Scrabble lingo (I'm not a huge fan of using this word as a clue because it's not very common. We were thinking maybe we were supposed to take the stem form of words for a while). Also, Unistems also seems to be able to mean to add any number of blank tiles, but in this case we were only ever supposed to use +1.

Anyway, I wrote up a program to calculate which letter to add, and we noticed that they were all direction letters. One of my teammates then took these directions and traced out the path, and we noticed the long cycle. So we had the 24 letters on the 2D faces of the tesseract, and then progress stopped. We were pretty much convinced that we were supposed to use the letters in the order that they were drawn by the path (since it did each letter in turn), but apparently not.

Then this puzzle got backsolved. After hunt I was thinking about what I would do if I were going to work more on the puzzle and I came up with thinking about the 3D facets, which turned out to be correct, although I had the benefit of knowing that the answer was eight letters at that point. I'm not particularly happy that extracting from those facets requires random anagramming, but I don't think it's a particularly bad extraction either.

Overall, I prefer the 10,000 Puzzle Pyramid, but I was super happy to see this genre of puzzle continue, and hope it continues into the future.

## Scout (Meta)

Nothing really special about this puzzle: the mechanic is exactly what the flavor text implies. We had a couple independent efforts to figure out the proper answer order and another teammate figured it out before me, which was pretty expected. One amusing thing is that when I was trying to backsolve I forgot that we had SKYSCRAPER as an answer and backsolved to get it again.

## Marked Deck

At this point it was a decision between three physical puzzles. The deck of cards looked really pretty so I decided to work on that. We decided pretty quickly that the cutouts defined an ordering on the cards so we sorted based on that, and the writing on the side was clearly encoded, which naturally is the Solitaire cipher. From that point forward, a few amusing things happened:

• Manipulating the cut out deck was pretty cumbersome due to the thin strips, so I supplied two of the decks that I carry around in my backpack.
• Our first attempt at decoding got confused by the fact that you're not supposed to add anything to the keystream if you hit a joker, so I started going through and recording the entire deck state at each iteration. (We had INSHUFFLK)
• When my teammates were double checking their work, they apparently messed up somewhere around iteration two or three and my full log bailed them out.
• Once we finally got the right order to in shuffle, we mistranslated 8 as G for a while to get ROLLING AND instead of ROLLIN HAND. Oops.

## Don't Look

Straightforward braille+knitting puzzle. I'm confused by the solution, though, because the dots that are listed in the solution as "k" are not the normal braille for the letter "k", which would be the top left dot and the bottom left dot, and similarly for "p". The closest we found was that the "k" is used for "the" and the "p" is used for "with", but once we decided that it was giving us a knitting pattern we figured out that they should mean "knit" and "purl", although I haven't found anything that would indicate that that is standard.

## A Pub Crawl

To be honest I like the gimmick of having a puzzle that doesn't seem to have much information. I wasn't a huge fan of where this puzzle ended up going, though. Once we found the Untappd profiles, we asked for the game materials and got a zip file of images. I was hoping for an actual copy of the game, although I understand that would be too expensive to get for every team. Anyway, we tried playing through it and recording the alcohol and fortitude values when a player was eliminated, but presumably ended up making a mistake somewhere.

We were particularly perplexed by what we were supposed to do with the Barmaid commenting about the state of the game when the players left a location, and then the players continuing to say the name of cards. Like, did they leave already or not? We tried ignoring those subsequent cards, but the numbers didn't work out quite right in the end. Trying to fix the error would basically mean doing everything again and we didn't have much fun with it, so we ended up just backsolving it.

Reading the solution now, it appears that the comments after the Barmaid were taking in the later round. That's extraordinarily confusing, since the timestamps aren't visible on the main check in page, and you have to go to the detailed check in link for each one, or else they just say "21 Dec 17". It was a cute idea but I wasn't a fan of the execution.

## Deploy (Meta)

One of the perks of being an experienced solver is that you look at a grid like this and immediately recognize it as Akari, even without worrying about "flashlights" in the flavor text. The flavor text did help with the 1337-speak connection, and so then people independently tried to figure out the right order of answers. Sadly it again wasn't me.

## Flee (Metameta)

We're now at our last puzzle of the hunt, and it was a long one. We sent out people to the radio boxes at first when we only had 5 of the answers, and I tried to coordinate everyone via Slack. Unfortunately Codex was working on the boxes at the time our people went out, so we just kind of sat there while our scouts were trying to report what Codex was doing and saying/hearing. I think we got about 1/3 of the graph connections from that eavesdropping.

By the time we got control of the radio boxes, we had figured out the sixth answer and sent out the last person, so it didn't take too long to collect all the data. Sadly, one of the connections was broken (we double checked) but hunt HQ had heard that there were some issues and came by to make sure that our data was correct.

And then we get to the logic puzzle part. Finding the path was pretty straightforward, and then we wanted to write in the letters of the tools into the grid, but the path was 34 tiles long and we only had 33 letters! After literally everyone got angry at that, we were just stuck, and eventually started calling in hints (the first Buzzy Bucks we spent were actually at this time but on the Serenity puzzle in Sci-fi, just because we were curious what we were supposed to do).

Eventually I went to sleep and it was solved by the time I woke up, and the answer was that we needed to count the number of edges in each graph. I'm sure that if we had done that at any point earlier we would have solved the puzzle pretty quickly, and that was just a blind spot caused by our tiredness.

And that concludes my puzzle-by-puzzle experience of hunt! Like I said in my last post, I liked the puzzles a lot, although upon reflection there were more issues than I had noticed previously.

## Bonus: Brainstorm

So we unlocked Brainstorm at 18:35 on Friday. Sadly, the team log doesn't have when we called in the "THOROUGHLY CONFUSED" instruction, but it was pretty quickly, and we had all the five words at that point. But we had misunderstood what it meant to "combine" the "ideas", and so we were trying to extract from these words metameta style, until around noon on Saturday at which point a member of hunt HQ came by and said "If you call in those five words concatenated...you will be done."