Mystery Hunt 2023

This post is about the 2023 MIT Mystery Hunt. You can see the hunt website with most of the puzzles and solutions at

This past Monday was the third Monday in January which means that another Mystery Hunt has come and gone. For the first time since the COVID-19 pandemic started, hunt was held in-person on MIT’s campus this year. As you’ll see, I was absolutely delighted to be back.


In the opening skit, we announced that the theme of the hunt was the “Museum of Interesting Things.” We also announced that all puzzles in this year’s hunt had been written by an artificial intelligence named MATE that we had built, and that MATE was so good at writing puzzles that future Mystery Hunts would be written by MATE instead of the winning teams.

As teams explored the Museum, they eventually broke out into the Puzzle Factory—a place where all the puzzles in the hunt were produced and the true theme of this year’s hunt. Teams eventually reactivated four additional AI’s, solved their rounds of puzzles, and made their way back to the factory—located in teammate’s hunt HQ in the Bush Room, and produced a new puzzle of their own—the 2023 coin.

A cog in the machine (in the puzzle factory)

My team, teammate, won last year’s hunt which meant that we were tasked with writing and running this year’s. I knew when we won that I wouldn’t be able to take on a major role, but even still I overestimated how much time I could devote to hunt.

For one thing, I was so busy with my new job (creating a real AI that writes code) that I didn’t have much time to work on our puzzlehunt (involving a fake AI that writes puzzles). In retrospect, I also was a little burned out from puzzle-writing; for the life of me I could not come up with new ideas of puzzles I wanted to write, and the whole cycle of choosing a theme, writing metapuzzles, writing puzzles, testsolving, revising, post-prodding and fact-checking them just felt a little less novel this go-around.

It was an interesting experience going from someone in a leadership position in the 2021 hunt to being just a cog in the machine. I recall telling Brian (one of our directors) that weekly newsletters were a great way to keep less active members of the team engaged with hunt, only to find myself getting all of my information about hunt from these newsletters.

As much as I tried to pay attention to what was going on, it was only when I arrived in Boston this past week to help out in person that I fully understood the structure of what we’d built.

Running the hunt

Running the 2021 Mystery Hunt mostly consisted of sitting at my desk and typing furiously at my computer. Running the 2023 hunt was much different. The in-person nature of it added mind-boggling amounts of additional work and bureaucracy. Unburdened from having any qualifications that made me an important person who needed to stay in HQ and focus on teams’ progress, I helped out however I could (move 10 boxes from room 2-150 to room 10-105, run Think Fast for Setec in-person then run it for Palindrome virtually, do a last-minute factcheck the clickaround to account for last-minute changes to the website, postprod a fix for an erratum for Conglomerate, answer hint after hint after hint after hint). It was all honestly really, really fun.

It was a joy to just be back, roaming the halls of MIT that I knew like the back of my hand, seeing what changed and what stayed the same. One of the first things I did when I got to campus was testsolve Subterranean Secrets a runaround in the tunnels of MIT that was so well-written it made me almost emotional.

Coin to be found later

OK, let’s get to the topic on everyone’s mind: why was hunt so long? By Friday evening, it was already apparent to us that teams were falling behind our solve-time targets that we had laid out. From that point forwards, hunt-comm in HQ was laser-focused on doing whatever they could to nerf the back-half of the hunt easier so that hunt could end on time.

Previous hunts recommended team sizes of 75 people for the optimal hunt experience. This year we lowered that to 60. Why? Well, as we stated at wrap-up, we genuinely thought at one point that we were writing a smaller hunt. We eventually determined that this was not the case (I think after running some simulations?), but by that point we had already made the size recommendation.

Why did we think we were writing a smaller hunt? Well primarily because we were writing fewer puzzles. The 2022 hunt had 196 puzzles; we had around 120. And yes, I know, I know what you’re saying: puzzle count is not a good metric for determining hunt difficulty, and it’s true: hunts run long not because there are too many puzzles to solve, but because teams get stuck; stuck on hard puzzles and more importantly, stuck on hard metas. If our metas were straightforward and our puzzles on average easy, then yes I think 120 puzzles might have played a little fast, but they were not and they were not.

You see, teammate loves unconventional puzzles (internally we call them “illegal” puzzles). The gimmick of embedding a puzzle in the loading animation just too good for us to pass up, even if it did make some teams confused and think our site was broken. Indeed, part of the premise behind the four AIs teams released in the third act of the story was that each AI was rejected by “teammate” for writing illegal puzzles, puzzles whose answer is in a foreign language, metas arranged in a sequence that loops back on itself, puzzles with answers that include new lines and punctuation, puzzles only accessible by playing a video game.

Dan Katz hasn’t released part 2 of his recap post yet, but its easy to predict that the thesis will be that we should have just cut Act 3 out entirely. I’m here to say that we simply could never have done that; when we won hunt we wanted to use the opportunity to write rounds with bold and interesting ideas—ideas that came to life in Act 3.

This is not to say that we didn’t mess up, but just to say that the fix was not excising rounds, but instead making everything easier. Act 1 (the museum) should have been entirely fish puzzles (which would have also made the hunt friendlier to smaller, less experienced teams), Act 2 (the factory) had way too many puzzles with unnecessarily difficult extractions after the meat of the puzzle was over (Think Fast, Subterranean Secrets, and Quilting Squares come to mind), and Act 3 (the rogue AIs) should have had easier metas and fewer puzzles.

Of course, by the time we realized all this, it was too late to do any of that, so we did what we could. We lowered the threshold to open up Act 3 from four metas to three; we started releasing free answers (“answerbergs”) like candy, and we tried to make every single meta that hadn’t yet been unlocked significantly easier.

Did it work? Well kind of. We had to make one final set of nerfs around 2am on Monday to the remaining two unsolved metas (Conjuri’s Request and The Tower of Eye), but with that The Team Formerly Known as the Team Formerly Known as the Team Formerly Known as the Team Formerly Known as the Team Formerly Known as the Team to Be Named Later was finally able to finish (with Death & Mayhem not far behind).

Since campus was closed between 1am-6am, they could not come back and do the runaround for several hours, but we still wanted to send out the email announcing that there was a winner, without naming who they were. We settled on this language, which I think is very funny:


Closing thoughts

There will inevitably be a lot of criticism directed our way for messing up the length. Much of it (maybe even most) will be deserved. Some criticism is probably even healthy for hunt itself; teammate is a young team, but we’re old enough to have a few members who were around in 2013; the backlash to that hunt informed our response to this one: we acted earlier and more decisively than we would have otherwise, and I’m sure future writing teams will learn things from the experiences shared about this hunt.

But to any teammates reading this post let me say this: hunters have every right to complain about how this hunt went, but you also have the right not to feel too badly. The single most important thing a person can do to keep this tradition we all love alive is to help write and run it. Together we made sure that

As long as we all keep making sure that this happens, hunt will be fine.

See you all next year 🙂.

FWIW, I planned what I wanted to say in the above section on the flight back in order to head off what I envisioned as a major backlash, but overall people have been quite kind in their comments about hunt. We received some very supportive messages from hunters on Palindrome, Left Out, Rage, and others, and teammate’s spirits are high. We know everything didn’t go exactly as planned, but we’re proud of what we put out nonetheless.