Solution: 🔲

At first glance, this is a straightforward crossword puzzle featuring images as clues. However, we quickly run into two obstacles while trying to fill the crossword:

• Many of the clue answers are the wrong lengths for the spaces they fit in.
• The crossword cells do not accept letters as input. It does, however, accept some numbers and special characters.

Consider 8-Across, which is an image of Kim Kardashian, but only has space for 7 letters. It is checked by KNOCK, DNA, RAT, SUN, and MAN, which reveals an answer with the form KARN??N. Though KARDASHIAN does not quite fit, it does share a suspicious number of letters. Looking at substrings, DASH pops out -- the answer would fit if DASH were to be replaced by a single character (or symbol, in this case).

Many of the other entries also contain rebuses of symbols or numbers:

Filling the rest of the grid in Google sheets (or equivalent), we get:

The next key observation is that each of the symbol rebuses is checked with a unique letter, suggesting a sort of substitution cipher. This is further confirmed by the fact that the crossword on the site only accepts these symbols/numbers and nothing else.

Filling out the grid with the substitution cipher, we get the final solution to the puzzle:

Authors' Notes

Dan: We initially planned to do a puzzle based on morse. Then I mentioned this idea and we ran with it instead. "Words containing other words" feels like a decently common puzzle mechanic, but it's usually fun and this implementation fit very naturally into this round.

The grid/idea was not that constrained; the main constraints were:

• We needed some particular words (not a lot of words have COMMA in them)
• We wanted grid entries to be cluable with pictures (meaning mostly nouns/names or sometimes verbs, with other parts of speech likeRED being relatively rare)
• We wanted each letter to appear several times, to make it easy to figure out what letters were used in the grid.

There was a draft of this puzzle where, by complete coincidence, the central row spelled RECODE. After having no ideas for what to do with that that wasn't a red herring, we changed some entries to remove it. (There were other changes, but we hadECHO/ECHO/MUSTARDin a different place and with * crossing H, rather than the currentACAI/RATIONS/MASK)

Unfortunately, there is a somewhat hard area in the middle left with MAC/COLONEL/CREIGHTON; MAC's M is checked by : and its C is checked by 8, so it's practically unchecked, and the image isn't perfectly clear (we tried a lot of images and it seems just hard to clue). COLONELbeing hard to clue and being impossible to distinguish from its abbreviation COL was another issue.

: and 8 are both not in that many words, so this grid area was constrained enough to persist Across several testsolves. If all else fails, it's possible to get it by noticing M and C are the only-remaining non-rebus letters.

Relatedly, CARETAKER (a.k.a. ^AKER) looking sort of like it's maybe just TAKER is somewhat unfortunate.

The puzzle mostly runs on the browser, through a combination of code that looks scary and code that may actually have some security behind it. The explanation of exactly how it works is too long for this author's note (the most interesting part is how it checks random small subsets of the grid to check if the grid is 80% complete, because I spent two hours reading the BCH code wikipedia page and noped out). This did lead to the puzzle initially increasing build times substantially and us cutting down on some checks as a result (if you saw the percentage tick up from <80% to 81.3% or something with one letter input, this is why; we accepted a 1-in-300 chance of this happening, rather than 1-in-1-million).

Finally, as someone who doesn't think we had enough animal pictures in this hunt, I was happy to be able to fit some in this puzzle.