2019 MIT Mystery Hunt, Part 2: Metapuzzle Development

(This is a recap/review of the 2019 MIT Mystery Hunt, which happened this month. Puzzles and solutions can currently be found here. These posts will contain many spoilers.)

I was spoiled for the last couple of years by having a “buffer week” between Mystery Hunt weekend and the beginning of my semester; sadly, that was a temporary luxury, as this year and next year, my classes begin a couple of days after wrap-up. Interestingly, I usually get a cold right when the semester begins, but this year it came later; so apparently a week after Hunt is actually when my immune system goes on vacation.

When last we spoke, I was lamenting the fix I’d gotten myself into, having committed to overseeing the meta construction/testing process for a Hunt in which metapuzzles had to be linked in a way that made construction/testing a massive pain. Hooray!

Those who know the Mystery Hunt well (and honestly, if you don’t, none of this is going to make any sense anyway) will recall that the strongest Hunt teams finished Setec’s last Hunt, 2017’s Monsters et Manus, in a much shorter time than usual. A lot of hunters, including some on Setec, were perfectly happy with this, since it allowed many more teams to finish the Hunt. But a lot of us, myself included, preferred for the first finish to occur a bit later.

I felt that one cause of the quick finish was how we tested metapuzzles for 2017; since 2011, Mystery Hunt puzzles have generally been written and tested using a handed-down platform called Puzzletron. One of Puzzletron’s best features is that once a puzzle goes into testing, testers can claim a puzzle, solve it (or attempt to solve it), and give feedback without the need for a moderator. In 2017, we set up metapuzzle testing in a similar way, allowing team members to claim metapuzzles to test on their own at their convenience. This allowed us to test metas more efficiently, but it also lent itself to generating easier metas. A lot of testers would finish metas solo, and so much of the feedback was based on how approachable the metapuzzles were by one person working alone. This is very different from how Mystery Hunt teams approach metas, and so when our solo-solvable metas were thrown to actual Hunt teams of rabid wolves, they got torn apart.

As meta coordinator, my plan to combat this was to do all of our meta testing and construction in groups, which is how we built metapuzzles in the 2009 Zyzzlvaria Hunt (scour the solution pages for those and you’ll find a lot of credits for Meta Group A and Meta Group B). So we began by splitting the entire team into three groups, trying to balance experience, interest in testing, and interest in construction. We also tried to ensure that each group had a time of week they could regularly meet on Slack, though groups were encouraged to collaborate regularly between these meetings.

We started by having each group (helpfully named Apple, Orange, and Tomato) generate their own ideal list of A holidays and B holidays, where A holidays were well-known holidays likely to appear in the first part of the Hunt (which would be unlocked earliest and reveal the “midgame” if completed), and B holidays were less obvious holidays that would appear later. We then merged those lists into a larger list for everyone to work with. The list contained all the holidays we used, plus the Fourth of July (A), Leap Day (B), Mardi Gras (B), Talk Like A Pirate Day (B, became an event), and Your Birthday (B, became a cover story for the April Fool in midgame). On our original list, New Year’s was an A holiday and Presidents’ Day a B, and those ended up switching.

We then had a round of construction where teams were asked to focus on writing approachable A-A metas and more challenging B-B metas; I think we allowed A-B meta proposals at this time, but warned they wouldn’t get in unless they were very strong. Groups wrote metas, we tested each with another group, and then each was left as is, scrapped, or revised, before testing with another group. After doing this, I worked together with various editorial types to identify the pairs we were likely to keep, and we designed some potential maps / partial maps.

Then we we went back to the groups and repeated the process, aiming for pairs that were likely to link the partial maps we already had. Some time during the second or third of these cycles, we noticed one of the three groups was less active than the others, so we split them up and redistributed them to the other two. We still wanted to be able to test things more than once, so we had to split each of the (now two) meta groups into smaller groups for testing; as the leader of Meta Group Orange, I found myself sending testing e-mails about Subgroup Mandarin and Subgroup Clementine.

A lot of great metas got generated that were discarded either due to not fitting well in the map, or just not testing the way we wanted. TK shared a diagram during wrap-up (viewable at 6:46) that emphasizes how many metas got proposed, written, and/or tested, and hopefully it communicates how challenging it was to build a set of metas that fit together into an elegant map; in particular, I was very insistent on a symmetrical map so that teams would get a sense of how it was fitting together as they opened new nodes. Off the top of my head, there was a beautiful Valentine’s-Fourth of July meta that got dropped because it was too easy to solve (and then trivially backsolve) with very few answers, and a fun Thanksgiving-Mardi Gras meta that simply didn’t fit the map. In addition, a whole bunch of puzzles that made it into the Hunt were adapted from proposed metas; We See Thee Rise was once Thanksgiving-Arbor, and Chain of Commands, Picture Book, and Schematics (minus the logic puzzles that ended up being the meat of that puzzle) all began life as metas attached to Bloomsday.

This entry’s getting a bit long (and my head is getting increasingly filled with gunk as it gets late at night), so I’ll continue metapuzzle discussion next post with some behind the scenes details on the metas we actually used.