This has spoilers for MIT Mystery Hunt 2022. Spoilers are interspersed and will not be labeled ahead of time.

So, teammate won Mystery Hunt.

Since the winner of MIT Mystery Hunt has to write the next year’s hunt, things will be…interesting for the next year. I still need to decide what level of involvement I’ll have in that, but to be safe this post isn’t going to do any speculation about the future of Hunt.

This year was the first year where teammate didn’t do open signups. In previous years, teammate was around 60-80 people, and based on a team survey after Mystery Hunt 2021, people generally felt the team was growing too big. So for 2022, we did a closed roster of dedicated team members, which was defined as people who helped write Teammate Hunt, or had hunted with teammate in 2021 and at least one previous year. I think this is always a tricky thing to navigate, since the natural state of hunt teams is to grow larger over time, and people will always end up close to whatever line you draw.

We had 53 people this year, which puts us around the size of Left Out. And this year was a really close race. Left Out was basically tied with us all through Saturday, and Death & Mayhem was ahead of us until they slowed down around Sunday 4 AM. (We…kept going.)

I don’t think we made any big changes from last year with regards to Sheets norms or remote solving tools. Based on what Wei-Hwa said in the afterparty Discord, I think the reason teammate edged out Left Out was that we were more gung ho about going for the win, whereas Left Out was more uncertain and didn’t fully let go of the brakes until the end. More specifically, Left Out saved all 3 free answers until they were sure about things, hammering them all on Sci-Ficisco. We had used one free answer in New You City, and two free answers in Howtoona, based on our observation that it would be easy to backsolve after we finished the meta. That likely sped up our meta solves and let us make up for our slightly slower puzzle solving average (relative to Left Out and Death & Mayhem).

I’m not sure how this compares to other teams, but teammate has always had a decent number of people who really like the meta-level strategizing of how to get from the start of Hunt to the end as effectively as possible, and team policy has always been that it’s okay to hunt this way. This means that, for example, you never need to ask permission to backsolve, and every year we’ve had a puzzle that got sniped from someone who was about to finish their forward solve. Depending on viewpoint, this is either fine or horrifying. But, well, it does work, and it can be fun to improvise a good speedrun route.

Thoughts on Hunt

Good hunt.

Fun theming, many puns, solid puzzles, good art, and cool that there was an intermediate tier of accomplishment between the first round and Pen Station. I would have preferred the team interactions to not be recorded ahead of time, since I felt that made it easier to skip the story, but I realize that it’s logistically a lot harder to make that happen while also running New You City.

So still. Good hunt.


Our pre-hunt socials were mostly different groups of 2-4 people solving Star Rats, and I think literally every group backsolved The Rescuers. This probably says something about what puzzles we tend to start working on first. We also played some escape room games and other board games.

Like last year, we had a #conspiracies Discord channel, where I claimed that it would be important that MLK day was a full moon. Nope, that didn’t matter, but we did find a Scriptnotes episode that seemed too good to be true. It mentioned puzzles, palindromes, and “anagrams of MATE”, and a tweet made on January 17, which was MLK day this year. Based on that and the mention of crossword alignment charts, it really seemed like it was seeding puzzle content. It wasn’t, and in retrospect, the date of the podcast (2 weeks after winning Hunt) should have been a sign that it probably wasn’t important.

For this hunt, we did three optional in-person hubs, with one in Bay Area, one in Seattle, and one in Boston. This was a bit risky, but I went to the Bay Area one since I felt it was within my risk tolerance. No one was flying in, everyone attending had gotten a booster, masks required, it was in South Bay where people generally take COVID pretty seriously, and you had to show a negative rapid test day of. I’m not too sure how much the in-person hub helped solving, because we still needed to join voice chat to talk to people outside the Bay Area, but it definitely helped on the physical puzzles.

I had been playing through the Ace Attorney series prior to Hunt. Right now, I’m on the 2nd case of Apollo Justice, which features a noodle shop called Eldoon’s Noodles. The characters mention it’s an almost-palindrome in game, and even talk about “Team Meat” when you inspect it more closely, which added up to an eerie pre-hunt coincidence. And then the first round was called The Investigation! Please, try to convince me this wasn’t a work of the divine.

Ace Attorney

The Investigation

The Missing Piece - First puzzle I worked on in Hunt. I recognized the Palindrome name badge theming right away. I did find the 2009 Star Rats badge during hunt, which was a bit of a trip, but I figured there was no way Palindrome would make a puzzle in the first round depend on previous real-world badges. We never figured out what the years meant - after we counted the lanyard overlaps, we were pretty sure that was correct and figured we had just skipped a step somewhere.

Where the Wild Things Are - This was the first physical puzzle, and mostly got claimed by the Boston hub, but the Bay Area hub got the jigsaw puzzle, which ended up being one of the longer ones. With no patterned edges, pairing pieces up was pretty tough. We figured out all of the mechanics, but our data was incomplete and the meta got solved before we could patch it up.

The Ministry

Harold and the Purple Crayon - fun dataset, although I came in after most of the IDing was done and mostly threw around extraction ideas with other people. I don’t love that you’re actually supposed to throw out the ones that don’t fit, but it is neat to see the final step.

Oxford Children’s Dictionary - We pretty quickly decided that “one side will be regular definitions, and one side will be jank definitions”, but it took us a long time to determine exactly what form of jank it would be. Once we got a few examples, it was pretty fun, although we ended up skipping about half of the letters out of difficulty, solving from “??? budweiser? + ??? seashore? ???”.

The Talking Tree - Pretty cute puzzle, and somehow these diagrams were a lot easier to fill out by intuition than the ones in the similar puzzle from Silph Puzzle Hunt. I did think it was funny that one person solving had written a phonetics puzzle for Teammate Hunt. Also I just realized how relevant the title is, it’s literally talking syntax trees.

Dinotopia - I view puzzles as creating order out of chaos. To borrow an analogy from Alex Rosenthal’s talk, it is a coincidence that the number of piano keys matches the number of constellations, but once you know this coincidence exists, you kind of have to make a puzzle that pretends this coincidence is vitally important. Sometimes you have to do a bunch of work to make the contrived coincidence work, but sometimes the pieces are already there and it feels like you’re just discovering it.

In this sense, it’s really cool that the writing system referenced in this puzzle is inherently ambiguous if you arrange the symbols in the right way, and that parsing out readable text is a satisfying challenge. I mostly figured out a-has of what we were supposed to do and how to extract, then got lunch while watching other people do all the work, which is really a 10/10 solve experience, would recommend.

The Ministry - When we unlocked The Ministry, we had 25/25 feeders thanks to backsolves. We noticed “bit = binary” right away, and I mentioned that COLORFUL HEAD could be a predicate for “starts with ROYGBIV”, at which point we quickly inferred all 5 mechanics. Since this was the only puzzle we had left, I got to witness the terrifying sight of Sheets locusts descend on the 5 x 25 bit matrix. I think we filled out all the bits in 30 seconds. Very scary.

Fruit Around - We didn’t pre-solve the “hungry caterpillar” connection before The Ministry, but we had a few minutes between solving The Ministry and doing the interaction with Palindrome, and we guessed “bookwyrm = very hungry caterpillar” in that gap, so Fruit Around was pretty straightforward for us.

In general, the construction of The Ministry is pretty impressive. I didn’t even realize that the meta answers were also semantic descriptions of the bookwyrm until after we finished Hunt. There are a lot of layers of constraints going on and it’s nice that the circle closes and it all comes together.


I mostly missed this round, aside from…

Curious and Determined - We ended up having to backsolve this puzzle, since we didn’t see the way to map letters to each clue. However, the realization that “wait it’s these colors” was fantastic, and it was funny we said “The Shawshank Redemption isn’t really any color aside from blue and orange, in the same way that every movie poster is blue and orange. I guess the main character’s named Red…wait a second.”

Lake Eerie

Large-scale Anthropomorphism - after IDing a few of the animals, we dumped all of the animals into Google at once, and got search results for taikyoku shogi, which translates to “ultimate chess”, a terrifyingly complicated game played on a 36x36 board. We IDed the appropriate adjectives, guessed the Chinese (well, Japanese) numerals were cluing rows, and figured it was a taikyoku shogi chess problem from hell. That seemed like exactly the level of ridiculousness to expect from Mystery Hunt.

Taikyoku Shogi

Now, if you solved the puzzle yourself, you might have noticed one issue - the game that’s used is actually dai shogi, which is played on a 15x15 board. Most piece movement is the same, but notably, the king in taikyoku shogi is allowed to move up to 2 spaces, which made it significantly harder to decide on the best reply. When we got the king out of checkmate according to taikyoku rules, we figured we had made an error somewhere.

Interestingly, whether you interpret it as taikyoku shogi or dai shogi, you extract FIC first, and we said “it’s probably something like FICTION or FICTIONAL or FICKLE” many times. We tried the first two, came back after someone told us we were using the wrong game, then got stuck with weird letters from a non-optimal mate where the falcon didn’t capture anything. After a closer reading of the rules, we figured out the igui rule to take a piece without moving out of the pin, extracted a K, and realized it was FICKLE an hour after we talked about guessing it. On one hand, we could have saved a lot of grief, but we also found the most unique part of the shogi logic, so I’m not too upset in the end.

The Graveyard - If I remember correctly, this was a meta that was only unlocked partway through the round. I took a look when it opened, stared a bit, tried to think of appropriately eerie connections, then said “hang on, aren’t these the Pacman ghost colors?” I confirmed the year lined up, someone else found the Japanese ghost names, and then we went on a backsolve spree. It was a little fuzzy, I believe we ended up solving from ROFA, TIONS, and a penciled in IRAPP on the group we had 3/4 on. I do think it’s a little odd that the Ghostbusters group was ordered by credits, when the names appear after 1/2/3/4 letters of the string.

The Quest Coast

A Number of Games - When this unlocked, I figured I would have to work on it because I’d done some combinatorial game theory before. But then it turned out teammate is just a bunch of nerds, because lots of us had done combinatorial game theory before. I drifted off of this puzzle in favor for…

Something Command - I’m heavily biased, but this is my favorite puzzle from Hunt. We did the math on the Eldrazi one first, in case the indexing was based on how much over lethal you could get. After getting lethal exactly with 2 different lines (attacking with Nettle Drone or not), we believed the extraction would only be based on the missing card name, solving at 4/7 correct cards. I think the puzzle is doable if you don’t know MTG, but knowing MTG definitely made it faster, and it was cool that you could intuit your way to figuring out the missing card even if you didn’t exactly work out all the math.

Sorcery for Dummies - Cool puzzle - I mostly came in after all the individual letters were IDed, to try to figure out paths for each monster. This ended up overlapping with technical difficulties that took down interactive puzzles though, and during the downtime I started on another puzzle. Once the backend went back-up, I decided I’d rather finish what I started and this puzzle was completed by the time I looked at it again.

Once Upon a Time in the Quest - When I went to bed around 3 AM Saturday morning, we hadn’t made much progress on this meta. I woke up at 7 AM, earlier than I planned to. I was going to go back to sleep, but I saw a message asking about Dinotopia’s mechanics. The overnight crew had broken into the Quest meta. I got on voice chat, described Dinotopia, then went back to sleep for an hour. Waking up to more forward solves in Quest, I read over the work, saw that THE DARK hadn’t been done yet, considered clue phrases that needed to start “THE DARK” instead of just “DARK”, proposed “The Dark Knight”, and was pleasantly surprised when that was correct. That got us to 6/8 of the Step 3s, which was enough to wheel of fortune the answer. I still think it’s a bit weird that there’s an extra IT’S at the start, but maybe it makes it easier to find good words or get to a nice round length.

New You City

Does Any Kid Still Do This Anymore? - I did this as my last puzzle before sleeping for Saturday, so it was mostly about grinding trigrams in a daze. I got Antioch in the first one from Nutrimatic, and decided that was fake until we did a few more examples. I remember saying, “I can confirm that kids still take the BART.” That’s technically true, but the meaning of the title became clearer after we got the main a-ha.

Bad Beginnings - We didn’t get this until after we won Hunt, but I’m mentioning it here anyways because the dataset used is great and it’s worth revisiting.

Proof by Induction - I came in at the end, after all the ideas were figured out, to help ID some of the missing lines. I hadn’t heard of this language before and it’s a pretty good one.


Sunday Dinner - Okay, I didn’t actually work on this puzzle, but I clutched the finale. We have a channel where people can crowdsource help on anything, and the call for help was “we have a cluephrase EIGHT PAST, about a NYT Sunday crossword with a food theming”. I tossed some terms into Google and said “SPAGHETTI”? It was right. I didn’t understand the clue at all. (They later explained it was an anagram and must have indexed a cryptic that used it before.)


Somehow I saw nothing in this round, not even the meta.


Rotten Little Scamps - I pitched in a bit to some of this puzzle, which sparked an ask for “does anyone have an Icelandic Nutrimatic”, which might be the best request I heard all Hunt. We did not find an Icelandic Nutrimatic.

Reference Point

You Took the Fifth - I got tagged to work on this puzzle the moment it opened. More of a word puzzle than an Ace Attorney puzzle, but still good, and it was cool once we figured out the reason it was presented as an It did take us quite a while to decide on the right interpretation of each line, but I believe that was part of the intended difficulty.

On Second Thought - Now that I think about it, I haven’t seen the phonetic step in very many puzzles before, but it didn’t feel too bad when we were working on this. The puzzle is kinda ISISy, but not in a bad way.

Diced Turkey Hash - Something Command was my favorite puzzle of the Hunt, but Diced Turkey Hash was by far the most memorable. This was the second physical puzzle, and I do think it’s a shame it unlocked so late. We figured out the binary grid pretty quickly, as well as the Mayan numerals and Tarot symbols. A bit more work got us the Dzonghka numbers as well. From there, we figured that “face-to-face” was a clue indicating we should take a walk between adjacent faces of the d20s. Based on the given text, we started identifying mappings between pairs of dice, noticing more relevant text every time we reread the given text. It’s impressively dense, and the “director’s chair” realization was great.

We struggled with it a lot, but given it was one of the least solved puzzles in Hunt, it wasn’t just us. Our difficulties came from two rabbit holes. First, the flavor about going on a journey really felt like we needed to trace a path along both pairs of dice, but the final extraction only relies on the mapping between the two dice, which could be determined without looking at the face topology. The given text about the journey definitely helped us confirm we were doing the right thing, but we also managed to translate it into a reasonably unique path that visited each face once. The most extreme argument was when we did the Tarot dice. We aimed for a path through all prime numbers in descending order, taking the shortest path when possible. At one point, the two primes were on completely opposite faces, and there was no unique shortest path, but I argued that “be a walker, not a rider - but do not wait” could be interpreted as “take the shortest path that does not go through The Chariot”, which left only one path on the d20. This was not the intention of that text, but it did lead us to writing down the correct pairing (via a circuitous route).

The other rabbit hole is something Palindrome definitely did not predict. We had rolled the dice, and didn’t spot anything weird in what face it landed on, but then we decided to try rolling them into water, which would magnify any weight difference. To our surprise, there was a consistent face that would point up. Given this didn’t show up at all when rolling onto a table, I was not a believer, until I tried for myself. I’ve recreated a video below.

I was still not convinced, and argued that other d20s might show the same behavior, even if they were fair. One person drove home and back to bring control group d20s, while we recruited help from remote solvers.

Discord screenshot

All the control group d20s landed on different faces between trials. And thus our solve group was split in twain - we could not deny the dice were weighted, but did it matter to the puzzle? Or was it just a manufacturing defect?

Arguments for it mattering:

  • This could be a reason this was a physical puzzle.
  • The dice feel high quality, which is evidence against manufacturing defects.
  • Previous Mystery Hunt puzzles have used gimmicked dice before.
  • The given text mentioned “an ocean of blue” - perhaps this was a hint to use water to figure out the dice were weighted.

Arguments against:

  • The distribution of faces when rolled onto a table is not clearly weighted (tested by doing the eye test of rolling the dice many many times). It would be very easy to miss the weighting unless you thought of the test we did.
  • Nothing about the puzzle presentation suggested the die would be weighted.
  • The gimmicked dice from the previous Mystery Hunt puzzle were more clearly gimmicked than our dice.

No one was making much progress convincing anybody else, so we decided to settle it by seeing whether the virtual version of Diced Turkey Hash left any way for a virtual solver to learn the dice were weighted. Only problem was that it wouldn’t unlock for another hour. So, we spun wheels a bit, until the virtual version unlocked and we learned that no, the virtual version only showed the faces of the each dice, and the weighting did not matter in the slightest.

It wasn’t real, it certainly wasn’t intentional, but it was a very entertaining argument, so thanks Palindrome, and thanks to the sponsor HRT for manufacturing slightly unfair dice.

Reference Desk - We random anagrammed the answer, and just could not figure out the ordering step when trying to backsolve. Without the ordering idea, backsolving was pretty impossible, so we just left the round incomplete. This likely contributed to our lower total solve count in the end.


How to Install a Handle - The moment this unlocked, I scrolled down the page, and said “YOOOOO IT’S MATHDANIEL SQUIRREL”. I had actually emailed the relevant dataset to Ryan North a week before Hunt thanks to this Dinosaur Comics, so it was fresh in my mind. I can confirm that he didn’t know about the bracket before, but does now.

How to Make the Right Move - The pair of us that first looked at this puzzle immediately figured out the puzzle mashup, then tagged other people to work on it. I credit working on the Sleeping Beauty meta from Inception Hunt. I also believe we sent an errata request about board 9, claiming it was impossible. It wasn’t, that board was just too bigbrain. We got it eventually.

How to Require Some More Assembly (Picture Puzzle 3): I got to come in to save extraction, which is always fun. Saw a completed grid and a highlighted entry, interpreted it slightly differently than the group that filled out the grid, and then we solved. Although it took us a while because we read “invert Y” as “flip across Y-axis”, which made the final image look more like a vampire bat than the actual answer.

How to Do Quality Reviews - Helped with initial data entry, but then it hit the part of the puzzle that was not very parallelizable and I drifted back to…

How to Find a Component - I opened the meta at the start of the round, didn’t see how to do anything, and left. A while later, I opened our meta sheet again, and saw someone had written “these numbers are the alphabetical permutation of the given words”. Seeing that all our answers were 9 letters long with unique letters, I wrote up my theories for how I wanted the meta to work, then tried starting from 3/9. We didn’t have enough pieces to place anything, but I suspected the mechanic was very constraining on the answers. I was right. After applying the constraints that the answer was 9 letters, all letters were unique, and numbers in each column should be unique, my Scrabble dictionary was reduced to just 226 words.

A teammate and I spent the next 30 minutes looking for thematic backsolves on abandoned puzzles, and failed to backsolve anything. In retrospect, most of the puzzle answers were not very thematic, and I suspect this was intentional. The constraints are really strong if you squeeze them for everything you can.

Around this time, we were down to 1 manuscrip left, and win-comm (the group watching overall strategy) proposed using the free answer in Howtoona. We had gone up to 5/9 Howtoona answers, and our estimate was that we’d need 1 more answer to crack the meta open. However, multiple puzzles in Howtoona were moving forward without getting stuck, and if one of those got finished, we could instead have an extra feeder in Sci-Fi. We knew that teammate was in contention to win (since we had gotten a phone call saying so), but we also knew we were not in the lead on unlock progress (because one puzzle had an errata issued 30 minutes before we unlocked it). After some discussion, we said to give us 15-20 minutes to solve the meta, then ask again. About 20 minutes later, we were still at 5/9, increasingly confident we’d need a 6th answer, and unwilling to backsolve for it because our candidate word list was too big. Given that, we pulled the trigger and redeemed our manuscrip on a Howtoona puzzle no one was looking at…right as the 6th feeder was forward solved. At 7/9 it was pretty trivial to backsolve the remaining answers and we finished the meta within 5 minutes.

I think the win-Hunt play would have been to use the manuscrip 20 minutes before we did, and the max-fun play would have been to use it on a Sci-Fi puzzle and let people finish their Howtoona puzzles. Instead we did something in-between that was somewhat unsatisfying on both ends. It wasn’t a perfect call, but it was a tricky decision given that teammate agreed we were both going to have fun and go for the win. I think the one we came to was acceptable.

In case you were wondering, that previous section is what I mean by “meta-level strategizing”. A conversation like that tends to happens every year.

The Plot Device

(This is wildly non-chronological, because I worked on this throughout the Hunt, but this felt like the approximate time where I did the most work on it.)

Narnia Beeswax - We got our first Plot Device solve in the middle of this puzzle. My understanding is that the group working on Order of Apparitions was getting really fed up with the “cluephrase” they had, and tried guessing the whole thing in frustration. In Narnia, we confirmed that we could submit single answers, and good thing too - it took us so many tries to solve EQUALLY SNUG. In general, the answer guess timeout was more annoying in this round, and we often spent 5 minutes locked out from guesses.

A Crying Shamus - I was not good enough to help on any of the cryptics, but I was good enough at searching all the words with “mystery” and “detective” to figure out the answer ordering. I was also good enough at reading last letters to finish the puzzle. For whatever reason the last letters popped out for me more than the first ones.

Synonym Toast - Just a neat word puzzle in general, although we ran into an errata on enumeration. I think we explained the issue to Palindrome quite poorly, because it took a few iterations for us to explain what we thought the error was. Sorry!

Step by Step Ladder - We had tried to presolve the Plot Device meta by building a word ladder, so when we unlocked the puzzle that was actually a word ladder, we sped through it pretty quickly (10 minutes based on our solve info).


I missed everything from this round aside from the meta, but I’ve heard good things about both Replicator Droid and Lists of Large Integers

Communicating With the Aliens - When we got to this puzzle, it was about 3:30 AM on the West Coast, and I was listening to ideas about lines of symmetry without really having much input. I felt I needed some sleep but also knew we were really close to finishing, so I decided to take a 30 minute nap and hope we’d have 1 or 2 more feeders by the time I woke up. I set one alarm (and 5 backup alarms), woke up at 4 AM, physically got out of bed at 4:20 AM (nice), and tried again. It took us an embarrassingly long time to reorder our puzzles in the given round order, and after doing so it also took a while to get just 1 of our answers in the grid. We had a lot of competing theories about what the symbols meant, none of which let us place the DES MOINES answer. Eventually, we got it in, and once we noticed how the three-dot symbols lined up, the rest was pretty fast. I decided to start the Nutrimatic query while most people were placing words, and mentioned a lot of the regexes were matching phrases starting with FIRST, which oriented us to the right pun idea.


We entered endgame knowing that the coin hadn’t been found yet, but we also knew we were probably very close with other teams, given we were behind on puzzle unlocks at one point. Our best hope was to assume we had leapfrogged teams on the metas and could finish endgame before they caught up.

Battery Pack - Unlike other teams, we did not pre-solve the Plot Device meta, but it was pretty clear what to do once we saw the shape on the meta page. We complicated it for ourselves a bit by assuming it would mix-and-match across all answers, rather than coming pre-grouped.

The Tollbooth - It took us about 1.5 hours to finish this puzzle, and I’m really curious what the solve time was for other finishing teams. We split into three groups: one to find books, one to solve book titles, and one to solve the printer’s devilry clues. We got most of the data, and assumed the extraction would be based on the pairing the puzzles described with the printer’s devilry puzzle. When we failed to notice anything special from that pairing, we tried increasingly conspiratorial ideas, including “take random letters out of each word to spell a punny phrase”, which got surprisingly far on building a reasonable yet totally incorrect answer.

With the entire Hunt as a potential data source, we tried a bunch of things, like the number of lines on each PDF, or the map on the Pen Station round page. After what felt like an eternity, someone who decided “borders” was important noticed the leaves on the PDF. We did the indexing, and won Hunt.

Personally, it didn’t really feel real. For me, it felt like, “oh, we won, wooooo, I’m going to sleep.” Even now, Mystery Hunt 2023 feels like this thing that doesn’t exist yet, and it’s hard for me to have an opinion on it before it becomes more solid.

Maybe I’m a little jaded, but winning Mystery Hunt didn’t light any big fires of motivation for me. I’ve already exorcised all my puzzle demons and written the puzzles I had to write. The ones left in my puzzle ideas file are generally uninspiring and don’t feel good enough for Mystery Hunt. After writing puzzles fairly continuously for 3 years (MLP: Puzzles are Magic into Teammate Hunt 2020 into Teammate Hunt 2021), I have a better sense of how easy it is for me to let puzzles consume all my free time and destroy my ability to make small talk or start any other projects that need concentrated effort. Sure, making puzzles is rewarding, but lots of things are rewarding, and I feel I need to set stricter boundaries on the time I allocate to this way of life - boundaries that are likely to get pushed the hardest by working on Mystery Hunt of all things.

The opportunity to write for Mystery Hunt doesn’t come around very often, which makes me think I should go for it, but I’m not expecting to write anything super crazy. Hunt is Hunt, and I am cautiously optimistic that I have enough experience with the weight of expectations to get through the writing process okay.