2018 MIT Mystery Hunt

We cruised through the Control Room without too much difficulty. The first puzzle I worked on, which was also our first submission and first solve (although I actually don’t remember the part where we got the answer, for some reason) was In Memoriam; other puzzles I worked on included We Are All Afraid to Die, Unfortunate Al, and Clueless. (Our second submitted answer was AD NAUSEUM for Yeah, but It Didn’t Work!; it didn’t work. Also, our team forward-solved Warm and Fuzzy.) I was responsible for solving the Fear meta at 15:48 with 5 of 8 answers, which was the Trendsetters’ first metapuzzle. I was pretty happy about it; it also enabled us to backsolve What The…, and maybe Roadside America to some extent (my memory is fuzzy on this).

I also worked on the Anger meta. “I love Braille puzzles!” I didn’t have the right usage of the words at first, and it took us a while to realize we wanted to use RAGE instead of ANGER, but the answer was nice.

We unlocked the Games island 16:54, which immediately became a bottleneck. I think the crossword referenced by This Year’s Hardest Crossword was not yet available, and Sports Radio and Flattery Will Get You Nowhere just both took time. We unlocked Sci-Fi just 18 minutes afterwards and tore much more readily into its puzzles, starting with Potlines and shortly afterwards Fuch-sia. The latter puzzle unlocked Lest You Be, which ensnared more than a dozen of our solvers. I had nothing to do with any of this because I was focused on Flattery with a bunch of other word puzzle enthusiasts, which became our first Games solve at 18:14. Afterwards I chipped away a little at Cartography but didn’t manage to do much and didn’t see it through. At this point I went to sleep for the night. Meanwhile, the rest of our team finished up the Control Room, then slowly worked their way in from the west, but we had many Sci-Fi metas and had unlocked Pokémon Island (in which we solved 33 RPM before Shoal Patrol) before even reaching the Desert. This finally happened early Saturday morning, when All the Right Angles fell at 03:04; at almost exactly the same time, we solved our fifth Sci-Fi meta (of six).

As the morning continued, I woke up to see Shoal Patrol get solved at 04:04, revealing Twitch Plays Mystery Hunt, which something like ten solvers jumped into. The short description of this puzzle is that we all controlled a ninja in a top-down 2D adventure game and could all issue commands to make the ninja move, Twitch-Plays-Pokémon style. We had to guide the ninja through guards that would look in various directions, ogres that would chase the ninja, and lava with changing safe areas, all in real time. However, each individual player was rate-limited to one command every 4 seconds, which was not enough to navigate the changing hazards, so collaboration was necessary to solve the puzzle (or, in theory, you could cheese it if you wrote some code to enable a central player to distribute commands to many clients, but I think doing it with live collaboration was manageable and more fun anyway.) Near the end they elected a coxswain and wrote detailed plans on the blackboard in order to get through. It was fun to watch. The timing of our unlock, in the wee hours of the morning when hunt HQ was sparsely populated, was unfortunate for letting people experience it, but probably helped us solve it efficiently without too much trolling. It felt like forever, but apparently it only took 100 minutes before we were ready to extract. We didn’t succeed at extraction though. More on this at the end.

Meanwhile, the Games round struggle continued. We unlocked GRATIA PLENA at 06:04 and Pestered at 08:48; the former would take more than 24 hours and the latter would be our last purchase of the hunt.

A bit later in the morning, I attended the Covering Your Ass with Caution Tape event. We had to make ’70s urban evening-wear with tape and caution tape, which I had absolutely no understanding of, but I had fun anyway.

Pokémon was still going much better; we unlocked You Know What’s Missing at 06:32, which gave us Mass Aid at 10:00, which gave us Harsh Financial Scrutiny at 11:00 — a great string of amusingly themed puzzles. After noon another solver and I WoFed The Taxonomist meta with I think three answers and a probable character class, which let us sidesolve Twitch Plays Mystery Hunt. It was time to play Under Control! This was the other highlight: we had to send one team member to a room and then, by watching him through a YouTube stream against a background he couldn’t see, direct him to make specific poses to battle a ninja master by typing into the YouTube chat. He received the commands slowly as what we typed was fed through a voice synthesizer and had to figure out what poses we wanted him to make, while at our hunt HQ, we had to figure out descriptions and labels for each of the ninja moves, as well as the pattern of which ninja moves beat which other ninja moves. So we did all that, had a ton of fun, and got back a scroll that… we also couldn’t extract from. See the bottom of this post.

The last Sci-Fi meta (Blue Sun etc) and the Advertiser meta both fell around 14:30. A bit later we finally got This Year’s Hardest Crossword, a forward-solve. (At some point while working on thie puzzle and making extremely slow progress through the hunt, we searched and found random crossword people leaving comments online saying it was a bit easy, which was both amusing and frustrating at the same time. We didn’t stumble across any solutions.) Then we solved The Desert. Probably the most embarrassing thing I did this hunt was trying to Wheel-of-Fortune this meta with half the answers (??T????A?T?LEM?IRE), calling in “OUTER COASTAL EMPIRE”, and then giving up. Somebody other than me called in the correct answer two minutes after I did.

16:20: We solved Starship Enterprise, the Sci-Fi metameta.

17:12: After six hours of toiling, we got Harsh Financial Scrutiny, probably my second favorite puzzle of the hunt. This got us to Hacking Island.

18:08: I helped break into the Pokémon metameta with Rapidash. I’ve been waiting for this moment — when knowing the lyrics to We Didn’t Start the Fire by heart would be helpful — for a long time. I think we figured out the general mechanism and Persian, Lycanroc, and Under Control (a backsolve that gave us quite mixed emotions) all quickly followed, getting us access to our Rival. Shortly afterwards, in response to solving The Lurking Horror II: The Lurkening, we unlocked The 10,000 Puzzle Tesseract, which I worked on.

I went to sleep for Saturday night a bit later; we solved the Rival and the Hacking Scout meta, and then backsolved the 10,000 Puzzle Tesseract, while I was out. It looks like the first extraction step from GRATIA PLENA also occurred in this time, at 03:45, more than 20 hours after we had unlocked it.

After I woke up on Sunday, I helped solve Blocked Lines (06:38), followed by Don’t Look (07:27). The latter was a remarkably unremarkable hunt puzzle — it was amusing that it was in a round of physical puzzles and we could go to HQ and look at the physical scarf but we could also just look at the picture. Probably HQ didn’t want to knit a hundred of these things. I wouldn’t blame them one bit.

Then at 07:45, we WoFed the Build meta. We noticed the words on the exterior of the answers, but entirely missed the “nut”s and “bolts” connection. Not expecting anything good to result, I tried putting each word in three times. Although it wasn’t super satisfying, it was mildly motivated by having counted the frequencies and realizing that every enumeration happened a multiple of three times, a leap somewhat inspired by missing Haddock Walk from the 2016 hunt. Miraculously, with maybe four letters and some probable character classes, we guessed the answer on our third try. I went on the runaround. It was cold.

Around this time, our team made the call to purchase Thanks with 100,000 bitcoin Buzzy Bucks. We had gotten “HER NAME IS” from somewhere but were not close to an answer and decided we needed more puzzles on Games Island.

After returning from the runaround, I helped find the second extraction step to GRATIA PLENA and put it out of its misery at 09:35, after more than a full day (a little ironic, come to think of it?) I think the puzzle really demonstrates the importance of counting everything. Discovering the fact that there were 365 words cast everything in a new light. Afterwards, I joined the group working on Death from Aslant. We never hit the intended solution to this puzzle because we never thought of treating the uppercase text literally, despite the fact that we also couldn’t find a meaningful interpretation of it; after I had stopped working on it, somebody else solved it by taking the set of accepted letters and anagramming with a wildcard. (With the setting and all the flags, DIRT OVAL would have been so thematic!)

Around noon, we got Deploy. It was pretty clear to us what to do in this one, although we missed the first letters of the puzzles and just tried slotting in the words in various positions until we got a unique solution from which we could read nice letters. We had enough people working on this; I mostly just contributed the observation that the “Elite” in the flavortext meant that letters were to be interpreted as numbers via 1337speak rather than A=1. With the Deploy answer in hand, we of course started backsolving. I also went on this runaround to take a break. It was still cold.

Again around the time I was on the runaround, our team decided to purchase America’s Best Friends, also a blocking Games Island puzzle, as our second purchase of the hunt for similar reasons.

At 13:45 we finally solved The Robber. From then on is primarily the story of our entire team getting stuck on Flee for the last six hours before we finished the hunt. We had drawn a path and gotten five or so independent verifications (we’ve finally been burned enough times by inadequately verified early-stage solutions), and tried assigning letters to symbols along the path from regular answers and meta answers in dozens of ways to no avail.

During a break, I looked at the Scouts meta, uncertain if we would need to solve it to finish the hunt, and managed to solve it. In hindsight I feel like we really should have noticed the awkwardness of the titles earlier. Still, I may not have gotten it if we hadn’t already thought for a long time about the badges from the flavortext and copied all of them into our solving spreadsheet. A bit later, some others solved Go! (and, nearly instantaneously afterwards, Dusclops) also during a break from this puzzle. At this point, aside from the Flee meta, our only remaining unsolved puzzles were Model Kit, Pestered, and Feedback Encounters of the CLD Kind. We made a lot of jokes that our IQ wasn’t high enough to understand Rick & Morty.

Still, most of the six hours was spent staring agonizingly at the Flee meta. Our numbers began to dwindle in frustration. (One person watched the entirety of Hack, Punt, Tool in this time.)

Having asked Life & Order whether we’d be able to spend Buzzy Bucks on metapuzzles and being told there was no ETA, we spent the rest of our Buzzy Bucks for the answer to Model Kit even though we were 90% sure we knew the graph node that would be unveiled. It was as we expected, but L&O started allowing yes-no questions on metapuzzles about an hour later, which was quite the frustrating turn of events. Eventually we submitted a long, detailed help request (saying, among other things, that our “morale [was] flagging”) and got two Life & Order members to visit. They were still fairly cagey about offering help, which I was a little unhappy about, but they offered to answer some yes-no questions for free, so we asked (1) whether the puzzle had the exact same solution if we replaced the symbols with arbitrary “enum values” (yes) and (2) whether the puzzle used the regular puzzle answers (no). I did not expect the answer to the second question, and don’t think I would have gotten this metameta without it. I was convinced we had to relate the symbols to regular puzzles to spell something out on the grid. This might have been caused by not-fully-consciously expecting a much more tightly constructed metameta because it seemed to be our last puzzle of the entire hunt, even though it might not be everybody else’s, and also knowing that every other metameta used regular puzzle answers.

We kept thinking, but it still wasn’t obvious what to do. As people left for food trips of varying length, I decided I wasn’t that hungry and stayed in our hunt room, where I tried to get into this zen revisualization mode to figure out what information we hadn’t used (I like treating my life as an Ace Attorney game, sue me). We had a meta answer, a digraph on symbols, and a grid of symbols. We could apply the digraph to the grid to get a graph on the grid; had we lost any information while doing so? I stared at the radio digraph on our blackboard, wondered how else I could use its information, wondered how many edges it had, realized that it matched the length of the corresponding meta answer (which had all been ingrained in all of our heads by that time), frantically looked up the other two graphs, and got the trick. So I guess that was pretty exhilarating and it might have been less so if we had gotten a bigger hint, but I don’t know. I feel like the ranking of teams after the first or maybe first three or so doesn’t matter too much; I would prioritize getting teams to enjoy the hunt after that point, and I suspect a nudge to, say, ignore the grid at first would have saved us a lot of anguish without making the realization less fun, but I can’t be sure and maybe others have different opinions. Regardless, I suppose the lesson is (again): when you’re stuck, count everything.

Some people retrieved the memory orb while I celebrated with a few others at Oath Pizza. I enjoyed my Luau pizza. Then we went on the grand finale. “It’s Spaceteam!” We finished (fifth) and did not need the music theory reference.

I think the only chronologically remaining thing to talk about is Pestered, which we purchased at 09:25 Monday as literally the last solve of our hunt. We had enough Homestuck experts that most of the realizations, connecting the speech to the characters and then to DNA and the zodiac signs, came easily. I have zero Homestuck knowledge, but I helped get past the first extraction (we had gotten the page numbers and indexed by the mod-100 remainders to get a lot of gibberish, but we hadn’t indexed by the first two digits; I was vaguely motivated by something we got stuck on Time Suck, from (again) the 2016 hunt. We still got stuck on the very last step, which was a little dissatisfying. I think at that point there was just slightly too much information (including the bizarre horoscopes in the just-published Tech that week, which we had been waiting to use all hunt!), which was mildly compounded by the somewhat misleading fact that the 12 zodiac symbols were unique and suggested a reorder (of course, the original reordering wasn’t alphabetized and could understandably have been significant, but then we had to make the opposite call in the Under Control extraction — see below…) and were also perfect for indexing into length-12 names.

At least I know now that Homestuck canon material is full of puzzlehunt bait. I might get around to it some day.

But now to return to the big ones:

Twitch Plays Mystery Hunt and Under Control

So again, solving these puzzles was extremely fun. Participating in Under Control was easily the best experience of solving a puzzle in any puzzlehunt I can think of; I’m not saying this lightly. Huge props to the organizers for coming up with them and running them for so many teams. But we didn’t manage to extract either of their puzzles and I think it really prevented them from achieving their full potential of how fun they could have been.

The extraction methods just weren’t natural enough to stand out among the huge range of other possible red herrings. For Twitch Plays, we tried to get more letters from the paths the person could take through the grid or get information from indexing into the victory messages we received after finding items. In particular, we (correctly) identified the ending message with bolded phrases as suspicious —

You may have evaded my guards, traversed my dungeons, and stolen my treasures [emphasis in original], but you will still need the password in order to confront me in another puzzle!

— but I think we only tried treating them as phrases to index into; we never came close to their correct interpretation as pointers to where to get letters, or saw that two of them could be interpreted to correspond to the six letters we already had. I also believe there was nothing to hint at semaphore and the intended order of the letters was not natural enough to let us pick up the scent of the right track. In hindsight, of course, these messages are supposed to point you to extracting letters from the guards, the dungeons, and the treasures, but then the bolded verbs are still solid red herrings.

The final Under Control scroll that we were supposed to extract from was similarly troublesome (the solution on the site makes it seem more straightforward than it is):

• The scroll’s flavor was full of red herrings (“love”, “heart”, “time”) that we thought could be relevant because they were so specific or so different from how the ninja master interacted with us during the event. We wondered if we were missing a reference to some body of canon knowledge involving a ninja.
• There was also lots of text that the ninja master said during the final boss that we tried extracting from; we ended up with a full transcription of every action and piece of dialogue from the final battle on our sheet that we did various things to, with no success.
• The order we had to perform the moves in the second round differed after our first failure, so it seemed to be randomized, and we didn’t particularly trust that the order we performed them against the final boss was fixed so that we could extract from it. I thought the more natural order, which was also more obviously fixed and deterministic, was the order in which the moves beat each other; of course, this presents the problem of there not being a natural starting point, but there are mechanisms that could have worked around that. Furthermore, since the moves on the scroll were the mirrored versions that were the versions performed by the ninja master, when we did choose to order by the final battle we focused a lot more on the order in which the ninja master performed them instead.
• Perhaps the most damning thing for me that would have been the simplest to fix was that the moves weren’t alphabetized, so we were quite a bit more reluctant to ignore their original order. I put a lot of effort into squinting at the poses in the order they were listed on the scroll and interpreting them as letters or numbers. After we had backsolved Under Control, our best guess for the solution method was still “squint really, really hard at the poses.” The first pose in the scroll did sort of look like an R, and both poses in the third column looked like Ms.

Overall, although the intended extraction was simple, I still thought it was unfair. There was just too much other information on the scroll and in the video that could be combined in too many other ways.