/dev/joe's Experience at MIT Mystery Hunt 2023

I could tell you I didn't want to post on the site owned by Russians, nor on the site owned by Chinese, nor on the site owned by a lunatic, but the truth is I wanted to post on the site where I've made spoiler blocks work. You'll see them in the sections where I describe puzzles, hiding any blatant spoilers; mild spoilers are unhidden. After all, to some extent the structure of this Hunt beyond the museum is a spoiler, and putting everything from Saturday evening onward in one big spoiler box is rather silly.

Sorry, this is long. Long Hunt was long and so is its recap. Use the headings to help you keep your place.

The links in this article (except where it is obvious they go to something outside the 2023 Hunt) go to the live Hunt site which is still live as I am writing this. If you are not logged in, you will get a 404 not found, but you can follow links from there to log in as a public access team with access to all the puzzles, and then the links will work.

The Tim Ticket

A couple days before Hunt, I tried to set up this TIM Ticket thing we learned we would need to get access to campus, figuring that as one of the people with the best knowledge of campus on my team, I was likely to go in at some point even if only to help others find things. Web site, create profile, assert vaccination status. A teammate had forwarded me the link above, which let me register an account, but I didn't have a ticket. Where's the QR code it wants me to scan to get one? Searching through the Museum of Interesting Things site (pre-Hunt) I found a link about the TIM Ticket which had a code in the URL which granted the ticket. Great, I got my ticket. Now I need to install the app that is going to let me use it when I am out and about. Got it. Installed it. Logged in through it. And then it tells me my phone won't work to open doors on campus because it needs NFC.

NFC is a technology I haven't had in a phone in quite a few years. In fact, I think only the Galaxy Nexus that I won for solving a Playfair cipher faster than anybody else actually had it. That was the only phone Samsung made as part of Google's "Pure Android" program, before they decided they were too good for Android and decided to try making their own heavily incompatible version, which landed them on my personal blacklist. When I needed a replacement, I bought whatever other Nexus was available then, and I bought other replacement phones since then, but what all these other phones had in common was the US versions didn't include NFC because it was deemed a technology that never took off here. Apparently all iPhones have it, and a few high-end Android phones, and quite a few if you bought international versions (which might give you greater ability to use your phone in other countries, at the expense of being unable to use certain cell phone frequency bands here), but I didn't have one of those.

The app offered me an option to print a card at a kiosk on campus, and since it was open without the card until Friday at 7, I decided that once I got my stuff set up at Le Méridien, but before Hunt started, I was going to walk down to the student center (W20) and get a card. The timing worked out to have me doing this during the 40 minutes between the end of the opening skit and the first release of puzzles.

During the skit, we learned (no big surprise) that the museum didn't have enough staff to open to the public, they were looking to hire some of us to fix that, and they were going to test our qualifications with puzzles. With that over, I set out for the student center. It was a short walk down Mass Ave to campus. On the way, I passed the famous Ire-Proof Rage Warehouse. The relevant lettering was covered by scaffolding, but it appeared they were repainting the letters. The ones on the long side of the building appear freshly painted. So I guess we are going to have this landmark for many years to come. It has been this way for longer than I realized. But pretty soon I was at the student center, and I had to ask somebody where the card-printing machine was. I found it. There was a screen with big words on it: Out of order.

W35, the Z center (as students call it), one of three connected buildings which make up the indoor athletic facilities at MIT, also had one, and it was right next door, though with the entrance on the far side. So I went in there, asked the student manning the front desk where the machine was, and she pointed to it... "Oh, it's out of order. There's another one in the student center."

"Yeah, that one's out of order too," I replied, and walked out, muttering, "Dammit, MIT."

On the list of kiosk locations I'd checked online before coming to MIT, there were two more locations I thought within reasonable reach in 16 and 32, so I headed across Mass Ave, down the infinite corridor, and down the stairs into 16. Building 16 is not a very big building. It's basically an open space at one end, a hallway with classrooms, and another open area where it connects to 56. So when I didn't see it in either open space I asked some student who was using a card at the vending machines, and he didn't know. (I found this one only after wrapup; it's literally under the stairs going into building 8.)

So I went to Stata, having decided that if I didn't get it here I was going to abandon the effort and just go solve puzzles. And after finding a vendor who had no customers at the moment and asking my question, he told me it was "down there" by the elevator, with the kind of bent arm gesture needed to indicate that it was way around the bend in this infamously nonrectangular building. But I followed the main corridor until I found it. Scanned my code. Some error came up about accessing the network. Was this why the other ones were out of order, or were those just empty from all the other hunters who did this before me? Anyway, I gave it a second try and this time it worked and I got a card printed up with my own face on it, which was clearly marked to show I was a VISITOR. I put the card in the nametag/card holder/pencil holder thing I've been using for all sorts of events since I got it at Gathering for Gardner 11, 9 years ago. I headed back to HQ, where I arrived only a couple minutes after the first puzzles came out.


We only had this round at first, so I dog-piled with the rest of my team into the few puzzles open.

Museum Rules was a good puzzle for the many eager solvers present at the start to dog-pile onto, as it consisted of a storybook with about 20 pages worth of pictures and a little text which we wrote down every detail about while trying to understand what the idea was. There are a lot of animals... but not all the pages have animals. Somebody latched onto the text matching the descriptions of state quarters, or state seals, but some of them didn't work, and then ...

somebody keyed into the idea that text on the last page clued supreme courts and discovered that there were seals for the state supreme courts, which sometimes differed significantly from the regular state seals. Finally, with some of the weirder but more identifiable pictures like using an elephant to plow a field, people did the right searches to figure out each depicted a weird law of some state

... which helped us confirm the states we were supposed to be using instead of just guessing or scanning for the relevant seal bits to match the puzzle text.

Inscription was a big mess of cross-references, with some visible, non-referenced bits that correctly led us to assume they were cryptic crossword clues. (This is a good puzzle for people who enjoy cryptics.)

Seeing how each clue had one of only a few references in parentheses at the end led us to guess those were enumerations for the cryptic clue answers. So we guessed those few words would be numbers, and clue 9, which referenced itself as an enumeration, had to be FOUR, which famously in English is the only number with as many letters as its value. Assuming the clue answers were in alphabetical order let us guess the rest of the numbers, and the clues for the numbers let us start filling in other words. Once we got several of the words that went in the main clue, that gave us part of the answer and we were able to guess the remaining part.

It was sometime around this point, while helping print a puzzle for someone, that I noticed I was staring at the loading screen for longer and longer times with each puzzle looked at, and commented on it to a teammate near me. "I don't know what sort of technology they ran this web site on. How can a puzzle which is just one screen of text (I forget which one it was) take so much longer to load than the loading screen itself? This turned out to be a key observation, as soon thereafter my teammates figured out ...

the loading screen itself was a puzzle, spelling out STOP, and entering STOP in the answer box opened up another round for us.

Several of us worked on Showcase, which was one of way too many puzzles in this hunt that consisted of a bunch of minipuzzles. I solved two of the minis while my teammates did the others, with a few proving stubborn and requiring teaming up to solve. We noticed that many of the puzzles had odd wording in the instructions or unnecessary comments, often but not always at the end, but it took longer than it should have ...

for us to extract the message from the instructions. Once we did, it didn't take long for somebody to find the programming contest which it referred to. And it was immediately obvious the weird answers we had come up with for the minipuzzles provided inputs for the problems.

But at this point I was alerted to us having the meta, The Cafe, available and all but two answers in the round, and I shifted to work on this. It was clear it was a kind of path puzzle, and one of my teammates identified the specific puzzle type as Anglers. In The Cafe (and in Anglers), you're filling the grid with paths, each path connecting one of the people (fish in Anglers) to one of the seats (anglers in Anglers, but normally represented as numbers outside the grid rather than seats on the edge). And one key difference is that in Anglers, the length numbers are at the edge of the grid (the seats), while in The Cafe they are at the people. That turned out not to make a big deal in practice, though.

I printed out three copies of the grid for those working on it, keeping one for myself. Others were working through our spreadsheet to identify and correlate all the things referenced in the puzzle and feeder answers. So pretty soon we had turned this into an Anglers puzzle with two lengths missing due to two missing answers, which corresponded to the people in the third row and third-from-bottom row in the grid.

The puzzle contained names and pictures of various dishes. They were references to real restaurant chains - the style of food matched, the pictures used the colors of their logos, and the dishes matched in various ways the names of the chains. The feeder answers each contained a word from a menu item at one of the chains which I assume was meant to be unique from across all their menus, which allowed matching them up, and furthermore, matching those answers to the miniaturized version of the dish images seen in the grid.

I assumed, based on the way Anglers worked, that we wanted to put a letter from the feeder answers on every square, which left our two unknown answers needing to sum to 27. This made a solution mostly along the lines of the one given on the Hunt site possible, despite lacking two of the lengths, as they are two of the last three to be placed. When I got around to trying to connect the missing one near the bottom, I realized it had to go by the shortest path possible because to do otherwise would mean the length-19 one would eat up too much space and prevent the other one from making its connection at all. We did the extraction, solved the puzzle, and provided the backsolve info (answer lengths and two placed letters in each answer) to the teams working on the remaining two puzzles, who got them quickly.

Other Early Rounds

By this point we had also solved many of the puzzles in Science, and I never actually worked on the round, apart from looking at the meta Nuclear Words briefly and making some observations other people had already made.

We were starting to get into both Natural History, Art, and the Factory Floor, and the next puzzle I worked on, Collage from the Art round, was another good dog-pile puzzle, a Funny-Farm-style word association puzzle with a big web of words.

The next puzzle I worked on was the first I did on the Factory Floor, Baking Bread.

Shortly after somebody figured out that the answers were spoonerisms of (sometimes) more reasonable phrases, they got that the title was ... almost a spoonerism of Breaking Bad. More correct to say it's a non-phonetic, letter-based version of a spoonerism. I didn't watch the show, so I let teammates figure out which character did each thing, and in cases they also confirmed odd-seeming answers were actual plot points within the show. I contributed that the recipe step icons corresponded to how you had to swap the initial sounds. (In the three-word answers, sometimes the middle word was unchanged and sometimes the initial syllables rotated through all three words.) Knowing this helped figure out more of the spoonerisms. I did end up doing some Googling about the show at the end for the plot points nobody could remember. I also tried to apply the money as offsets of the characters' initials, but when that didn't work we went to actors instead. That was still wrong, and I moved onto another puzzle before my teammates figured out the right way to apply the money.

After this, I worked on the Cute Cats puzzle from the Art round a bit, though I had many teammates helping to figure out the three distinct types of answers in this puzzle and what to do with each one. It fell quickly, and I moved on to Catenaverbozoa from Natural History, which, despite the name, had nothing to do with cats.

The name should be parsed as CATENA-, chain or connected series, -VERBO-, word, and -ZOA, animal, which describes what we were doing. Each fill-in-the-blank clue can be solved by combining a prefix, infix, and suffix from the lists of Greek and Latin word parts provided, to form a pseudo-scientific name for a creature which literally does whatever its name suggests. For example, if there was a brontosaurus (thunder lizard) in this puzzle, it would literally create thunder.

We needed to translate these names into English names, and anagram the English to match phrases describing the bizarre pictures at the bottom. And we actually only had a few of these completely right at first pass, but it was enough to convince us we were doing the right thing. It was a slog that took a long time to complete most of the answers, and when we had a few left in the bottom half, a teammate anagrammed the letters we extracted from the top half without the bottom half clue (which was going to be a corresponding pseudo-scientific name) to just slightly short-circuit the puzzle.

I looked at Quality Assurance from the Factory Floor round. My teammates had already done the first step, solving the logic puzzle normally, ignoring the provided but erroneous solution, and I started on the process of verifying the assumptions in the erroneous solution. But I was getting too tired to keep working on it, and at 11-something I took a look at the big board to see our progress.

We have a "big board" with origins back in the Beginners' Luck team, which consists of a bunch of laminated cards we can write puzzle names and answers on and stick on the wall with masking tape, for tracking open puzzles, solutions, meta solving, etc. We have an electronic version of this as well, which was invaluable the last two years, but we still maintain the physical board.

On the big board I saw we had two metas solved and several answers in a few other rounds. It wasn't the level of progress I expect from my team on the first day, but I felt it was the puzzles that were difficult, and not our own performance, and so I felt like this was going to be a long Hunt.

I did one last thing before bed, which was to go into campus with a teammate to pick up two physical puzzles. I tried my new snazzy VISITOR card at the door to Lobby 7. The strip on the card reader turned green, but the door did not unlock on the first try. Then somebody came out, so just went in. Weaver was a bunch of strips with letters and numbers on them that clearly needed to be woven, and Tissues was indeed a box of slips of thin paper, but they had drawings of bodily tissues on them. I was too tired to help with either, though.

Saturday Morning

We solved, as it turned out, 8 puzzles between midnight and 5:30 when I returned to hunting. And during the first couple hours I didn't solve any more, mostly jumping around to look at puzzles opened overnight and not getting ideas. We had figured out that G|R|E|A|T|W|H|A|L|E|S|O|N|G was a loosely connected Morse crossword ...

and the answers were going to be entered as Morse. We also figured out the clues were given in order of answer length (in Morse, counting spaces) and then Morse-symbol-alphabetically, for a particular sorting of the Morse symbols. An additional complication was the clues were given as a single continuous string of Morse; we had to infer where the breaks between clues fell. I helped by writing a Morse dictionary. I wrote all 2 and 3 letter combos as "words" due to the number of abbreviations we'd already found to be used, and the 4s from a word list, written as Morse and sorted in the same way, to better allow figuring out missing words. That would be enough to get all the answers up to length 11 or so as long as they weren't made mostly of E's and T's.

We ended up not solving this one before its meta was solved, though, and it became a lesser priority.

And then I worked on the cross-number puzzle Hall in Lost To Time.

I pointed out that this was written in the style of a puzzle from the 1995 Mystery Hunt... which I then looked up and discovered was a puzzle missing from the archive of that Hunt. Given the title, the others should be old, lost Hunt puzzles as well, which certainly made sense for the antique image of Netscape, though I didn't recognize any others. I stuck with it until we almost finished the cross-number puzzle, utilizing the fact that the current year was 1995, but still having some difficulty in the lower right because, at this time, we missed that there was an erratum in the 1995 Hunt archives we could apply here.

I stopped work on this puzzle to begin a shift as Wrangler (a role responsible for adding puzzles into our tracking system, which makes spreadsheets available for them, etc.) I made sure the previous Wrangler was caught up, and then I was free to do some solving.

After those false starts, I finally contributed to a solution after I found Vraal working on This Puzzle is Just Another Regular Cryptic. His group had just finished the first stage of this puzzle, which gave the clue

NOW DROP EVEN WORDS and they had discovered that reading only the words in odd positions in each title gave new clues. Without actually working it out myself first, but noting that it had the same wordy feel as the clues, I said, "Apply it to the title." When we did, we saw it made "This is Another Cryptic." That wasn't necessary for solving, but I continued to work on the cryptic clues, which revealed a third level of the puzzle as a single cryptic clue which we solved, one of only four puzzles we solved during my two-hour shift.

Once that one was through, I looked at several other puzzles, sometimes making minor contributions, such as in Art of the game.such.fame! in Art, in which I reviewed the regular expression puzzles and corrected an error in one. At this point we had not yet figured out what the answers to these meant, but when we did, it didn't take long to finish the puzzle.

We did unlock a few puzzles during my shift for me to wrangle. After one of them, I gave Charles a thumbs up, explaining that I successfully wrangled a puzzle called into our solving system. At one time, this system was not Unicode-clean, but the usage of emoji characters in recent hunts made it clear we needed to fix that. (Aside: My Hunt Index had a small issue with that, too. After the 2021 Hunt, I discovered that my tools for maintaining the Hunt Index site were almost Unicode-clean. I use a tkinter front end to Python to enter and edit stuff. Any character I type or paste into the GUI works and makes it into Python. But when I try to edit a puzzle which has Unicode characters in it beyond the Basic Multilingual Plane (specifically, emojis in U+1XXXX), when my Python code tries to load those characters through the tk part of tkinter to display them, it breaks. It's a known limitation of tk's Unicode support, which only supports characters up to U+FFFF, and not something that's easy to fix. I worked around it by having Python check the entered data and replace any such characters with their HTML entity equivalents, which my code has always supported. Back in the aughts when I was developing the initial version of the index and I had no expectation that my software had proper Unicode support throughout, I was using é to write Pokémon and the like.)

I looked at Broken Wheel for a bit, filling in a few answers.

By the time I looked at Kubernetes, my teammates had gotten all the data into the spreadsheet, but there was a lot of solving left to do, to which I contributed a bunch. This one was based on the convention of abbreviating long tech words by writing their first and last letters and the number of intervening letters as digits, which I am most familiar with from I18N (internationalization) and L10N (localization), but which is also apparently used for Kubernetes as K8S. But they had done this for ALL the words in a bunch of clues.

The first level of this gave us categories of things with a specific order and the members of the category, which helped us put the words in the second part into the correct order. These words formed slogans or descriptions for various open-source tools, except that one of the abbreviated words in each description was superfluous. The superfluous "words" had the right first and last letters for the products, but the number was wrong. The numbers instead provided an ordering for this step. We made a couple mis-steps in trying to do this final extraction, including at one point interpreting "Who is this dev working for, anyway?" from the flavor text to mean we needed to find what developer or organization was responsible for each product, but when we realized the number of items in the ordered set matched the length of each product's name, we got on the right track.

Saturday Afternoon

We used some of the free answer tokens from events to unstick some metas, so we solved the Natural History meta while I was working on Kubernetes, and the Art one shortly after noon.

My first afternoon puzzle was Second-Rate Tiles, which nobody had started, so I worked on tediously extracting which letters were reused from one word to the next in the video, which was sometimes difficult due to the speed with which they were moved. When that was done I had some people helping to actually solve what the weird words were.

We got our break-in at CORTILE, which sounds like the clued word QUARTILE and fits the other associated clue about the Latin for heart. This let us determine the words we wanted were spelling bee words, specifically "losing words." The standard spelling bee rules require the last contestant remaining to first correctly spell the word his oppnent misspelled, and then correctly spell a new word, to become the champion; else the previous contestant who first misspelled a word remains in the bee. It is usually the "winning word," the final word in this sequence, which is cited in spelling bee writeups, but just is important is the "losing word," the one misspelled to begin the sequence. If you Google "CORTILE" and "misspelled," you find an article about the 1997 Scripps National Spelling Bee, in which CORTILE was the losing word. Suddenly, a lot in this puzzle made sense. The reference to 14-year-olds was based on the age limit for the event, and the definitions were all wrong because they were a bunch of weird words most people who haven't studied for the spelling bee don't know. We worked from various references to fill in one word after another. It took a while, but eventually we spelled out the phrase formed by the letters she put into the bag during the video.

When that was through I moved on to The Typesetting Machine, where they had solved most of the words and were working on a couple remaining ones.

I was interrupted about 3:00 when we made the discovery that the gizmos on the factory floor, which we had noticed earlier but not understood, ...

affected puzzles in the Innovation round. Specifically, each gizmo changed specific data in a couple puzzles. For example, there was a Maze with letters on the paths in which some of the paths and the entrance and exit moved and some of the letters changed depending on the settings of 3 gizmos. In a wrong configuration, there might be no maze solution, or there might be one or more solutions but the letters along the path were gibberish. If you submitted these gibberish answers, you got a response indicating the answer was correct, but something was wrong with the puzzle. The people working on them had been frustrated by seeing the puzzles occasionally change, and they had figured out random team members messing with the gizmos were the cause. So I had to make a team-wide announcement not to mess with the gizmos.

When I got back from this, they had solved The Typesetting Machine. I next spent some time assembling a chunk of the Quilting Squares grid. It was clear we wanted a criss-cross grid, with most entries having one or two colors making up most of the entry, a few being a variety of colors, and occasional off-color squares (perhaps at crossings?)

I stopped the quilting assembly when we solved the World History meta, which unlocked our first capstone (meta-meta) puzzle, MATE's META. And this is where one of the flaws of the web site really stuck out. You couldn't print any pages with the Factory Floor theme from the web site. Something in their page styling (I did not dig in to try to debug it) makes everything on the page disappear in print media, so you print a blank page. And even though this puzzle was a meta-meta for the metas in the Museum which did not have this theme, this puzzle had the factory theme. We'd worked around this on some other puzzles by opening up images we wanted and printing those, but the only way to print this one was to copy and paste the whole puzzle from the web page into another kind of document and print that. To have had one puzzle that didn't print properly could have been excused; to have a major part of the Hunt unable to print properly is a problem that surely should have been detected.

In MATE'S META, it didn't really take us too long to discover ...

how the meta answers clued ways to extract two letters from each answer, which assembled an 8-letter word in each row, and then per the instruction above the puzzle we had to split it somehow into two words. Before too long we figured how to do it using interleaved words with a word in each line being a rearranged version of a word from the previous line. I am not sure we thought about the permutations exactly the same way the posted solution describes, but we did something equivalent with the permutations to match them to the grid and extract the answer.

So now we could focus on more of the Factory Floor, Basement, and Office puzzles (and another group was working on the Hall of Innovation which all needed to be worked on together).

I looked at Gears, a Rows Garden variant, with the blooms being replaced by gears that could rotate into 6 different positions. But the fact that there was additional weirdness going on made this too difficult to figure out. We did get it without using one of the free answer tokens, but more than 24 hours later.

Saturday Evening

At dinnertime, after printing the grids in Reflective Screen from the Office the hard way (by opening each image in a new tab and printing it separately) I joined the group solving it. Over a couple hours, we painfully solved the first logic puzzle which normally doesn't have mirrors, but here it does. About this time, teammates solved the meta for this round and we gave up before trying the other 5 of these and a mini-meta.

I worked on X Marks the 🍎. Somebody had started this one, identifying four of the six names from their family trees in the Peach minipuzzle, and they'd written in a few of the states seen in the Sugar Maple minipuzzle and all the characters and their species in Red Maple. I quickly identified the grid for European Beech as a ...

tree of Morse code symbols, with the given letters disambiguating and helping me quickly find that the answer was MAIN. One of the possibilities for the Peach mini given the solved clues was SIDNEY, which led me to conclude they were streets near campus. And with some research I found the relevant diagram of the relationships of the mammals, of which the grid in the puzzle is the subset corresponding to these 8 species, and pulled AMESBURY from that.

They told us physical presence was needed, but it seemed likely what we were supposed to do was find the tree(s) of the given species on each street, draw an X, and THEN go find something at the crossing of the X. But who wants to go hunt for trees in the winter, in the snow, when they are hard to identify? I went to Street View, and to MIT's database of trees on campus. I found one maple on Amesbury Street, on the side that wasn't MIT property (so it wasn't in MIT's database) which I supposed could have been a red maple. It wasn't the variety which has leaves of a deep and very obvious red. The other streets were much longer and not as easily searchable. It turned out all the trees we needed were from parts of Cambridge that are not on the campus, and the city has its own public tree database. I don't think we ever found that, so we never found the trees, nor the intersection of the big X, which was indeed exactly what we were supposed to do.

I looked at Sliced Up several times, especially since a couple of our teammates chose to write all the words on mini Post-Its and put it in the middle of a table. It was clear we wanted to make clues out of the words, but how? We were misled by things like "atmosphere sailors dislike" formed reading around a corner in the top left. There were simply too many wrong ways to read sensible clues in these words. We never got it.

A little after 9 PM we finished the meta for the Basement round, which opened another capstone. Didn't we still have the Factory Floor and the Hall of Innovation? It seems Teammate made those optional to unlock this capstone, which was really important, as solving it unlocked the 4 remaining rounds in the Hunt. Weirdly, this one was not a meta-meta, but a self-contained puzzle. Maybe we were meant to unlock one drive mini for each meta solved? Whatever. We had Reactivation and dog-piled on the mini-puzzles it contained.

This puzzle was like the Reverse Dimension round from the Zyzzlvaria Hunt, which had us matching up Doctor Who doctors and companions to piece together puzzles that had been split into two parts. Here, part of each puzzle was on one of the AI drives, and part on one of the files given independently at the bottom.

I worked on two parts of this. First, there was a set of clues for a long, looped word ladder that we had to put in the right order. This went with a graphic that explained which way to move for a change in the first, second, third, or fourth letter (though in each case, two moves in opposite directions were possible), and additionally told us to fit the entire thing in an 8x6 rectangle and not to cross over our path at a diagonal. I finished the drawing at just about the same time as a teammate did so independently.

The other part I worked on was a set of minipuzzles which one of the files revealed were variants on Dungeons and Diagrams, a D&D-inspired logic puzzle with Battleships-style clues outside the grid but symbols inside representing walls, treasures, and monsters and a completely different set of rules regarding their placements. The file revealed that all the monsters in these puzzles were variants on the usual monster rules. Despite never having seen the puzzle type, I was the first to solve one of the grids, the one called "The Two Keys," and I helped correct some mistaken impressions about what some of the variant rules meant. The puzzle worked, but it was a weird design choice. Which was the worst part of this puzzle? That it had minipuzzles inside a minipuzzle, that it made up variants on an obscure puzzle type, or that it used figurative language to describe those variants?

I got too tired to stay awake to see the end of this, but finishing it allowed us to start opening the AI rounds, and it really wasn't possible for them to have opened the rounds without having us solve this puzzle first because the answer to this puzzle was the names of the four AIs, which were all over the other rounds, so it explains why they sped up access to the puzzle.

Sunday Morning

By the time I got up, we had just solved the second level of The Wyrmhole... But this round is complicated, so it needs a recap.

Much like ⊥IW.nano in the 2021 Hunt, this round has a nested structure where after solving a set of puzzles and a meta, the meta becomes a normal puzzle contributing to the next level. In the first level, the feeders were all copies of puzzles we had seen in earlier rounds, and we just had to re-enter the same answers from those puzzles. We also got from Teammate a set of triangular tiles with letters on each corner to use for the meta. The second level was normal new puzzles and a meta which did not use the tiles; these were just solved. So we had just started the third level.

The Devil's in the Details from this round had been identified as printer's devilry, but the answers were weird, and I noped it.

Specifically, the printer's devilry answers (the words that had been eliminated from the clues) were the names of demons from the Ars Goetia, but with a letter deleted from the demon name - a deletion within a deletion, effectively. So the strings you were looking for were doubly nonsensical. My team had actually figured this much out, but I still headed for saner territory.

I moved on to 5D Barred Diagramless with Multiverse Time Travel over in the Admiral Boötes round, because it sounded interesting and because my teammates had only barely started on it and those who did that little bit were probably now asleep. I watched the video about the chess game, which looked over-complicated, but I hoped I could figure it out. It seems like I needed to do the crossword first, anyway.

Crosswords, plural, that is. It was a Siamese triplets barred diagramless, and of course the bars were not in the same places. The "little bit" somebody had done already was only about the first three rows of one grid. I got all three grids started, and others joined in when they woke up and helped finish. (There are four more paragraphs of this discussion inside the spoiler block.)

We found some clues that contained what we at first thought were extra words. Later, we figured out they were alternate clues for words which were different by one letter. The different letters all fell into crossing words where they made plausible alternate answers to the clues at a time in the past and/or in some parallel universe. We came to understand that the three grids lived in different years, thus providing the timeline order, and each grid had two solutions which represented different timeline branches.

There were also crossword answers related to the chess itself. One of these confirmed there were two boards with movable pieces (indicating that there was only one live side-branch and the main timeline) and another clue spoke about "the" branch and its answer confirmed black made the branch. This was relevant so that we knew the position of the boards. The versions of the crossword solutions that correspond to our reality, past or present, are the main timeline, and the other version of each crossword solution lies on the branch, which was made by black, so it goes above the main timeline. Meanwhile, if we make a move through time, it will create a branch below the main timeline.

Ultimately, we solved all the crossword grids and started solving the chess. We understood the green squares in the last example represented checkmates. Another crossword answer told us we needed to make one move - a mate-in-one problem, which is great because trying to analyze this crazy game any deeper than that would pretty much be impossible. Between the crossword clue about boards with movable pieces and the 5D chess game rules, we understood that pieces on boards in the past cannot be moved, but kings in the past can be checked and checkmated. Such a check can only be avoided by capturing the attacking piece or, if the checked king is two or more boards in the past of the attacking piece and the attacking piece is not a knight, moving a piece onto an intervening past board in the path of the move. Except another crossword told us the only pieces black had were kings, so he couldn't do that either; he'd put that king into check doing so. So to escape a check of a past king, black could only capture the attacking piece on the board it was attacking from. (A third possibility was also eliminated by black only having kings. When you move into the past, the "present" moves back to the newly created board on the new time branch, one time step after the board the piece moved to, which would buy him move time to avoid being mated or maybe even allow him to move the past king. But kings can only move one step into the past, and their branch would be at the current present. It takes a longer move into the past to do this, which black is incapable of.)

We ultimately did not solve this puzzle, knowing from the puzzle we needed 22 mate-in-one moves and only having 4 of those. Most of the missing ones relied on a trick we didn't consider: When we make a branch by moving to a board in the past, all the other pieces which were on that past board come with it, and it's possible one of those white pieces will check a black king in the past of the branch. So any move to the board will create a checkmate. If we had figured that out, we probably would have finished the puzzle.

In summary, I think the crossword part of this puzzle is good fun puzzle you should try if you didn't do it (though it will take you some time), and I won't look down on you for skipping the chess part.

And I had another wrangler shift, so there were some moments I was pulled away to do that duty, but this was really the only puzzle I worked on in the morning, only to give it up.

Sunday Afternoon

I looked at Dispel the Bees. My teammates had gotten most of the words, and we knew what we were doing with them, but actually doing it was a challenge we were failing at.

It was the Spelling Bee puzzle from the New York Times, where you have to find words in a grid of 7 hexagons, reusing letters any number of times, and the center letter must be used, but in reverse. They gave us clues for 7+-letter words which would use all the letters, sometimes repeating some, and we were supposed to place them in the hexes. The clues were in groups of 1, 6, 12, 18, and 24 with appropriate repetition of letters so that we would build 6 Spelling Bee grids centered on the 6 outer hexes of the first grid to make a bigger hexagon, then 12 grids centered on the new outer hexes, and so forth.

Early this afternoon they started giving us "strong" answer tokens that could give answers to AI puzzles (the tokens we got from the events earlier could not) and told us we would get more hourly, and we ultimately decided to spend one of our free answer tokens on rather than continue to struggle with this.

Update: I went back to look at this one on 1/28/23. At the time we killed it, somebody had (maybe 10 minutes earlier) started a program to solve it. I wanted to see how long this would take.

I wrote Python3 code in about an hour, including debugging, which finds the grid solution in 8 seconds, given all the words. We didn't have all the words, but had all the first 7 words and were only missing one or two in each other ring. The program is written to accommodate this, and takes this strategy:
The center word is HEXAGON. Some of these letters are repeated in the second ring of hexes arond the center, but the H and A are not; each appears in exactly 3 of the 6 words, with two words having both. This corresponds to no H and A in the second ring of letters and H and A in the first ring adjacent. To eliminate rotations and reflections, pre-place these two letters adjacent in the first ring. This reduces the permutations of the other HEXAGON letters to 120.
Subsequently, it works one word at a time, trying it in all places in its ring. Since each placement (with the inner ring filled) covers four or five used letters, there are only 6 or 2 permutations of the new letters, so the branching factor is small, and many words won't fit in each location.
The numbers printed by my program are status checks. Each line tells how many grid possibilities it has when starting each word, which ring it is on, and how many words have already been placed in that ring. I left them in because they would have been useful in determining how hard it was going to be to solve the remaining answers. If we didn't have one or two answers, we'd get close to the end of one of the rings, and there would be (as it turns out) 2 to 4 solutions for the words we had, and we'd stop and see what we could make from the empty slots in these to match the unsolved clue, and move on.

Assuming the programmer was of comparable skill to me, and that we could solve the remaining words as quick as my analysis suggests, it would have been a bit over an hour. Did we waste our answer token? There were other puzzles we forward-solved after this point that took more time, and we didn't solve this meta for another 11 hours. So maybe, but it wasn't clearly a waste with the info we then had. It would have taken most of the hour to determine if the problem was tractable.

The next spoiler block contains the first line of my program, which defines wds as a list of all the clue answers in order, in all caps. Feel free to skip this unless you actually want to run my program yourself.


And this block contains the rest of the code.

import copy
import itertools
for j in range(21):
  for k in range(11):
    ltrs[-1].append(" ")

#directions that we can move to from a center
#forced cells to break reflections and rotations
#center cells to fill at each stage
for j in range(5):
  for c in rings[-2]:
    for d in dirs:
      if (nc not in rings[-1] and nc not in rings[-2] and
          (j==0 or nc not in rings[-3])):

#functions for letter banking and permuting words
def uniq(wd):
  return "".join(sorted(set(wd)))

#Place a word in the grid
def plac(grid,word,pos):
  #Collect the letters already placed
  if grid[r][c]==" ":
  elif grid[r][c] not in remain:
    return res
  for j in range(6):
    if grid[rr][cc]==" ":
    elif grid[rr][cc] not in remain:
      return res
  #Get permutations of remain
  for rp in rps:
    for j in range(len(avail)):
  return res

#Place first word
#Place words in other rings
for rn in range(1,5):
  for wd in range(6*rn):
    for ltrs in todo:
      #Place word anywhere in ring not used
      for pos in rings[rn]:
for ltrs in todo:
  for r in range(21):

I looked at Crack the Crypts at this point and again a bit later on. We had all but three of the cryptic clue solutions by the first time I looked at the puzzle and I added one more. My second look did not yield any further understanding; by that point we understood the common feature for each type of lock, but actually understanding how we were supposed to extract an answer from this puzzle took until just after midnight.

I looked at Space Modules, the meta for the Admiral Boötes round. It was pretty quickly clear we were overlapping the filters with some of the given letters in each constellation to make something, but others were getting them done faster than me so I moved on.

Vraal and I looked at Pluck the Petals, which some teammates had already started. They just needed some data cleaning, fixing up wrong answers, in order to bring the extracted phrase into something recognizable.

It was a bit of a pain that the Project Gutenberg version of the book referenced in this puzzle is messed up, with some missing and duplicated pages in the part we most wanted, the one which tells us the meaning for a given flower. The other copies of the book we could find just copied their busted one, but the text was searchable and we could often find the results in the reverse index (flower for a given meaning). Ultimately we got enough of the right pieces to read the extraction.

I looked at Investigate Relics and I had absolutely no idea what was going on here. Some of my teammates thought it was some sort of video game, but at no point did anybody guess anything remotely correct.

It was easy to be attracted to the Eat Desserts on Main puzzle. While there might be many places on Main to get dessert, veterans on the team knew a popular one we thought would work. I remembered the menu board having fewer columns and more rows than the puzzle, but the shortest away-team mission of the Hunt confirmed they had updated it since I was last there and it did match the puzzle now. This was a pretty quick solve after that, followed by a longer away-team mission because it was the "Bring us food" puzzle of this Hunt. Of course it was.

Sunday Evening

I finally got back to the Wyrmhole round, looking at the third-level meta Lost at Sea. My contribution was ...

pointing out that ships' hull numbers changed when they were reclassified, and some of the years only worked if r. meant reclassified, but it took several tries to get the group working on this to notice. I think because it was late in the hunt, some people were not as sharp by this time. I know I wasn't, despite being the one to make this particular observation.

When we got that solved, it opened a round where the puzzles were all blank. But the meta was a copy of that fun puzzle Collage from earlier, and teammates quickly figured out what was needed for these puzzles. And that opened Period of Wyrm, the final meta for this crazy multi-level round. We figured out pretty quickly we wanted to ...

match up pairs of answers to each occupation, with the names being there solely to provide an acrostic of MANDELBROT SET, which told us the math to use. Not too long after that, we realized one of each pair of matched answers was real and one imaginary (i.e. from a work of fiction). Some of them were pretty easy, like Old Hickory being a name for real US President Andrew Jackson, and Underwood being a fictional US President in House of Cards. Getting Laborer as referring to the Labors of Hercules was a bit tougher; even though we think of the labors as entirely fictional, the way one of the answers was written (apples) made it the "real" version.

I think the Indexer was the last one we got (or at least getting that one let all the possibilities we had for some of the others collapse to unique choices).

The Tower of Eye fell soon after that (I never worked on it), and that left just Conjuri's Request, which took place inside the video game Conjuri's Quest. Also, we had a team meeting. Despite the fact that it was almost Monday, we were apparently close to winning the Mystery Hunt. We knew we were capable of writing a hunt; more or less this team wrote the pretty successful 2018 Caltech Puzzle Hunt, and certain members of this team had been on several teams that wrote hunts in the decade prior to that. We affirmed we were ready to write another Mystery Hunt and went back to solving.

The round structure was ... less complicated than some of the other AI rounds in some ways, but different.

There were 14 puzzles, and they worked more like regular puzzles than in some of the AI rounds, with normal answers. But you couldn't check answers for them normally; there was no answer checker on the puzzle pages. You had to go into the game, where there were treasure chests on the map, each corresponding to one puzzle. If your team had not solved the puzzle, there was an answer blank there; otherwise it told you that your team had already solved the puzzle, and you could pick up a key. There were doors that required keys, but unlike usual video game keys that disappear when used, these didn't, but you needed a certain number of keys for each door. This might mean that you couldn't reach some puzzles' chests in some instances of the game, but the maps changed and you could find another where you could reach it. Once you solved enough puzzles, you could usually open the entire map.

There were also a bunch of monsters. Each monster proposed a puzzle challenge for you, like anagramming words, playing Wordle, solving a geographical trivia question, and many others, but always the same type for each monster type. Somewhere (usually behind a door requiring 9 keys) there was a boss. When you met the boss, you would get a long series of the same types of challenges the monsters gave. Once you beat the boss once, in any game instance, you got access to the spellbook (an option on the main screen, outside the actual dungeon map). This let you upgrade the items you could equip at the start of each dungeon instance so they were more powerful. To do this, you had to provide a chant, which was the result of applying one of the spell names to one of the puzzle answers. For instance, applying Time Slower to SECOND HAND provided MINUTE HAND.


Remember back on Friday when I said this was going to be a long hunt? They gave us 22 extra free answers (beyond the ones worked into the hunt structure as event rewards) and we are still here on Monday trying to finish it. The intuition from solving a lot of Mystery Hunts doesn't tell you how long or short it's going to be, but you definitely figure out early on whether the Hunt is more likely to end on Saturday or Monday.

We had three feeder puzzles left on Conjuri's round after using all those strong answer tokens, and we figured we didn't really need them. We had a clue that strongly suggested that ...

something in the number of times we had to fight each monster in the boss fight, along with the map, was relevant, and we were doing it entirely wrong. The first theory proposed was that we number each room as we enter it, take the numbers of rooms where we encountered monsters, add the number of times we fought that monster in the boss fight, turn that into letters, and it spells something, possibly several different somethings, since we knew we got different numbers of monsters in some game instances. The first mapping to track all this information started out looking good, and he thought maybe he'd just messed up counting. But we disproved this; those numbers weren't consistent, by monster or by position, not even if there was some way they mapped to the puzzles/spells. Which we really should use for something other than making the game easier.

Shortly after midnight, Tyler finally figured out Crack Some Crypts, and we were down to two. And then somebody (think it was Chris King) just outright guessed ...

what the Smoke Machine might upgrade by entering the upgraded chant in the spellbook before we had the corresponding answer. Yeah, that worked. And then it was easy to backsolve the puzzle by trying the obvious answer in the two remaining puzzles. Well, relatively easy, but we still had to navigate the map to find the chests.

Shortly after 1 AM Teammate sent out email saying meta hints were coming "before 2:30 AM." So we had three teams mapping instances of the game, using the feature that let you join a teammate's session, trying to collect as many fully mapped instances as we could. Some of the first mappings we did didn't include everything, but now we were collecting the location of every chest, every monster, the keys required to open every door, the order we visited the rooms, the details from the boss fight, everything. On each team, one person was controlling the game, the other mapping, communicating over a voice channel, so that the mapper could help the controller, for instance, not waste time going into previously explored areas. In some cases a third person was video recording the whole game to give us a way to go back and confirm data. And I was looking at the data collected, which continued to look completely inconsistent. Every theory we proposed was invalidated multiple times, showing it was not just a mapping mistake but a wrong theory.

It turned out the hints came out right at 2:30, hinting for Tower of Eye, Period of Wyrm, and Conjuri's Request. I guess none of the leading teams was stuck on Boötes. The hint for Conjuri's Request was exactly what we needed. It eliminated a bunch of possibilities for what we were supposed to be looking at. So it was less than an hour after the hint before we solved it.

This unlocked MATE'S Team, a third capstone puzzle. This was a sort of scavenger hunt through the entire Hunt, giving us a bunch of snippets of puzzles modified or presented in one of several ways, and each with a word or phrase redacted. We had to go back to and click on the relevant section of the original puzzle to "find" a jigsaw puzzle piece that went back into the capstone. So everybody awake was trying to recognize the bits from the puzzles they had worked on. The jigsaw solution suggested how to extract a letter from each redaction based on the way that puzzle snipped was modified.

This puzzle took us less than an hour, and we got a call from Teammate telling us we had a 6:30 AM appointment for the runaround in Lobby 7. First or second, surely, with the campus opening at 6 and Teammate surely needing a little time to set up. I took the time to pack up the stuff I brought to hunt, keeping just one bag I could carry with me on the runaround.

The runaround was thankfully short, just a bunch of fun little activities. Each activity was based on a meta answer from one of the AIs, for instance ...

"import antigravity" resulted in antigravity charades. This was regular charades but the people acting out clues also had to keep bouncing a balloon into the air while doing it. The wrapup included only a still picture from this activity, but in it (at this time stamp), I'm the person lying on the ground, doing the crab part of Crab Nebula.
For completing each activity, we collected an oversized gear that we used at the end to fix the puzzle-making machine in the Puzzle Factory (which was in Teammate's HQ), which had the coin in it all along.

Front of
      the MIT Mystery Hunt 2023 coin, showing the 5 AIs

Update: Wait, is that a puzzle? The lines at the bottom of the coin... (identified by David Latham 1/29/23 and solved by oneplusone an hour later)

Think of these as traces on a circuit board. The circles appear in the middle of the pattern represent holes where the trace goes to the other side, and in each case it crosses over one or more other traces that move over together and comes out on the other side of those traces.

This represents the permutation mapping 1234 to 4312, or more importantly, TEAM to MATE. The lines on the right are a doubly-flipped version of the ones on the left; they are flipped directionally (so that the far left maps to the far right and the middles map to each other) and by position (so the inside of the left lines maps to the outside of the rgith lines). This illustrates a symmetry of this particular permutation, which I've illustrated here by writing TEAM left to right across the top-outside ends and MATE in the middle. Writing TEAM in the reverse direction in the middle and MATE reversed on the outsides illustrates the symmetry in a different way.
Solution to puzzle on coin
Back of
      the MIT Mystery Hunt 2023 coin, showing the Puzzle Factory