A puzzle a day...

17 May, 2026

On Saturday I came across this mega-video on the Cracking the Cryptic Youtube channel, where over the course of 3+ hours (!) one of the hosts solves a puzzle posted by Jane Street (investment bank & prolific sponsor of nerdy Youtube). I watched about 15 minutes, but then stopped because I wanted to have a go myself, and managed to find a few hours (across 3 sessions) to solve it.

I then went back to watch the Youtube solve (yes, even I find it difficult to believe that I enjoyed this), and I was amazed to see that he did the whole thing without any external bookkeeping, just storing everything in his brain as he went (i.e. only entering guesses and deductions in the puzzle itself). It was mesmerizing to see him leap from one clue to the next (admittedly after about 30 minutes of getting oriented). By contrast, I poked at the edges a bit, but fairly quickly did some systematic analysis of the puzzle.

Briefly, the puzzle shows a network of 30 cities, each of which has a transformation rule (e.g. anagram, opposite). Ten "packages" consisting of a single word enter the network and transform according to the rules of each city they pass through until they arrive at their destination as another word. The trick is that only two of the cities and none of the paths (other than entry and exit) are labelled! I noted that the cities fell into 4 categories¹, depending on whether the transformations worked on the letters themselves or the meanings of the words:

1. Entry and exit lengths match exactly (e.g. anagram)
2. Entries are all one letter longer than exits (e.g. chop off the first letter)
3. Entries are all one letter shorter than exits (e.g. add a letter)
4. Unconstrained (e.g. opposites)

This meant that when looking at a given unlabelled city on the map, you only had to consider a subset of all the possible rules, rather than scan the whole list. I also printed off the top half of the puzzle so I could easily refer to the rules without scrolling away from the map, and kept a spreadsheet that tracked which cities I had already guessed from each of the categories. If you watch the Youtube solve, you can see a few places where this system would've been useful, though it is remarkable to me how small the effect is for him.

We had similar entry points to the puzzle, in that I also connected NOSE to London via guessing the correct intermediate, guessed two relationships further, and then got stuck at a fork (hard to say more without spoilers). Then I moved to a completely different spot to attack from the edge, and our solves went very differently. One thing they had in common though, is that neither of us had many false guesses at all. The puzzle writers did a magnificent job of making the right answers really feel right, even though there's not much explicit information. This is the mark of a good puzzle (thinking back across years of MIT Mystery Hunts), and very hard to do well, as you have to put yourself in the frame of mind of a solver who has no idea what's going on, whereas you can see the "obvious" correct answer. Kudos to Jane Street, and thanks to Simon from Cracking the Cryptic!