Thank HN: The puzzle game I posted here 6 weeks ago got licensed by The Atlantic
theatlantic.com
theatlantic.com
The game will stay free to play (and not require logging in). Also, I'm still making all the puzzles!
HN provided the first real infusion of players that weren't my mom's friends. So thanks everyone.
FWIW The Atlantic's team is amazing and got this live exactly 2 weeks from when we signed the deal.
This happened quick and I feel very lucky. The HN community of solvers keeps me honest with much helpful technical and editorial feedback. I love it all -- here or at mayor@bracket.city
T[Tom who befriended a volleyball] HN
PS my original post! https://news.ycombinator.com/item?id=43160542
On the other hand I like when I can ‘skip ahead’ and then try to reverse engineer a trickier inside clue.
I would say yes. It would feel great, like you are outsmarting the game. Kind of like when you discover a good build in an RPG that helps you get ahead - even though it's often intended by the designer, you feel like you pulled a fast one on the game and it feels awesome.
I see it as a UI limitation that you can't fill it in (how would you show it?). But it's fine as is, I'll just use it to help me solve the adjacent ones.
Edit: And congratulations!
"Go hang a salami, I'm a lasagna hog"
or
"A man, a plan, a canal, Panama"
though it seems that all of the answers (in the one tutorial and one round that I played) are single words, often slid in as character insertions. Just go light on the spacebar, apparently.
I've played about a dozen of these now, and very often I can see what one of the big outer clues is way before I've worked out the inner clues.
If I just solve the big outer clue, the game is done in 20 seconds or so, and I'm not sure I'd keep coming back.
Rather, the challenge is to use what I've understood to work out the inner clues. This is tricky because I've got to keep more in my head at the same time, Towers of Hanoi-style. Or I sometimes write it out.
I have my notes in an open notepad from yesterday, so I can reproduce my thinking without spoiling today's:
[scold, with "at"]: hmmm, no idea.
[something golfers' apparently [scold, with "at"]]: not sure...
["there[something golfers' apparently [scold, with "at"]]]: Ah, ok, got it: [there[X]] has to be "therefore." So I see [something golfers' apparently [X]] is going to be "fore."
So if "fore" = "Something golfers apparently [X]" and X is [scold, with "at"], I can see that innermost one must be "yell."
if not i agree with you. it'd be nice if it flashed green or something around the entire set of brackets but made you still answer all the sub-clues though.
"If you answer early, we'll ignore your guess without letting you on."
I'd really like to be able to just answer the puzzle without answering any of the intermediate stages. It is a much more challenging feat to just hold it all in your head and then type out the one long answer than to answer the individual stages. It promotes some real mental modelling skills that way.
The game, played as-is, is almost no challenge at all. It just feels like busywork.
Did they take your product as is in those two weeks or rebuild it?
One minor bit of feedback/request: maybe I'm too spoiled by code editors, but it would be nice to have a visual aid for identifying matching brackets -- maybe colorizing matching pairs ("rainbow brackets"), and/or a click-to-highlight feature that highlights the entire contents of a pair of braces for you. I felt like I was spending a lot of time trying to count bracket pairs, which made it hard to keep track of where I was in the puzzle and was less interesting than trying to solve the wordplay.
Amazing game!
Also, I would make it more obvious which clues are eligible for solving at the moment rather than penalizing us for not being able to discern which ones are.
Contra several other people here, I also like that you _can't_ skip ahead - yes, I know that's probably "Venus", which is a good clue for working back up to the clues I _can't_ figure out. It's the journey, not the destination.
(If you have any doubt whatsoever about whether the terms are confidential, assume they are, and don't put anything at risk.)
- Without the tutorial, it's confusing that you're not supposed to click and you're supposed to start typing. I wonder if placing the text box at the top would make that more clear.
- Some of the clues are confusing due to inconsistent punctuation. For example:
[to ___fish, to lure someone in using a fake internet persona] = cat
[do this or cut bait] = fish
[taking a pay one is a bummer] = cut
[rocks when added to soda will NOT cause your stomach to explode] = pop
> The first line uses a comma, the second line uses "or", the third and fourth lines don't have any punctuation at all, so the sentences make no sense.
... There is no inconsistency there. The 'or' and the comma in lines 1 and 2 are not parallel to each other; they're doing different things. Neither could be replaced by the other without changing the meaning of the clue.
Similarly, in line three, nothing in it could be replaced by a comma or by a disjunction. (But, and I want to emphasize this, line 2 doesn't even contain a disjunction; you appear to have misunderstood all of the clues.)
Line four is a bit different in that it contains a grammatical mistake. It should say [rocks that when added to soda will NOT cause your stomach to explode]. Other than that... it's a fourth style of clue. It isn't comparable to the other three, and there's still no inconsistency.
What do you imagine would add "consistency" to these clues? #s 1, 2, and 4 could be unified like so:
[to ____fish, to lure someone in using a fake internet persona]
[____ or cut bait, common idiom]
[____ rocks, rocks that when added to soda will NOT cause your stomach to explode]
I think my confusion with clue 2 was that I had never heard of the idiom "[to] fish or cut bait" [1]
Otherwise I quite liked the puzzle.
Glad to hear you will still be making the puzzles yourselves.
There is one point of feedback and it's that I would like to have a native keyboard on iOS instead of a virtual one. For the rest this game is great.
I realize it might be easier to write a puzzle than solve one. Writing them seems quite hard at first, but you have the solution and can choose words to break into clues, and from there do the same again until you have a properly complex puzzle.
Congratulations on getting it picked up by the Atlantic.
Did you approach Atlantic, or did they approach you?
I’ve had a word game in-mind for years, and several attempts to implement it. It’s a tricky one but I’ll crack it one day :)
My plan is to be acquired by a news company.
So, how did you go about this? I have some ideas myself
I feel like this is going about it the wrong way. This puzzle game licensed by The Atlantic for example, was made because they wanted to provide a fun game experience for their patrons.
With games, the best games come from an organic experience. Its almost worse to have a preset plan.
Not to mention, making a game is work - you have to have approachability, rules think through alot of scenarios even for "simple" games to make sure things make sense etc - and if you haven't even done it before I highly suggest you actually try and make a game first
But I had all the game logic and flow mapped out long before I thought about platform.
Yes games are tough to make, I’ve made a few so far and it’s always a struggle
My only nitpick: I totally get that you have to solve the inner clues before solving an outer clue, and that outer clues can help you solve an inner clue.
BUT, the ergonomics/UX of that is just impossible... There was an inner clue that was super-stumping me, and I could solve "downwards" from like 5 clues above, but trying to keep track of which brackets were which and which level was which was maddening.
It really feels like it needs a way to answer outer clues even if it won't collapse them. Just to keep track of them. And then a better way of tracking so many levels of nesting. I don't know if the answer is indented nesting like in code, or what. But the current UX only effectively allows you to work inside-out, when outside-in is clearly a necessary piece of the puzzle, because sometimes you get inner clues that are just stumpers that you have no other way of figuring out.
I loved the concept, just not some of the clues
The custom keyboard makes all the tools you need a first class component of the game.
But yes, it's currently a little clunky.
My one piece of feedback is to improve the keyboard (if that’s at all possible). I regularly miss points because of typos even if I had the correct answer.
I'd be very curious if you could share what that process looked like in general? Did they reach out to you, how did they find you?
Were they interested in the gameplay alone, or the player count / growth?
Was it much work, technically, to get integrated on their website?
And of course, how long does it take you generate one full puzzle?
- it would be nice to have a list of the incorrect guesses that you've made. Sometimes I come back after a while and can't remember if I already guessed something or if I just thought about guessing it but wasn't sure... Of course this also dovetails a bit with what some of the other people mentioned about guesses being counted as incorrect even when they're guessed early because then if you have a list of all earlier incorrect guesses then an earlier incorrect guess might now be correct, so I would also change that, that a guess which is only incorrect because it's not available yet isn't counted as incorrect, and instead it says something like "not yet!". Another related problem for me though was I got stuck once and it turned out I had just made a typo in my earlier guess and hadn't noticed, which is why it would be nice to see your earlier guesses. Oh, and also related, I think it counts multiple incorrect guesses of the same word (again not 100% sure since I couldn't see my earlier guesses)
- it would be nice to improve the stats, maybe give you some kind of distribution to see how you're doing, maybe on the calendar you could colour code to show what days you've done well and what days you did worse or something
One little nitpick is that while I prefer hardmode for puzzles like this, on mobile (where I prefer to play such games), it's WAY too easy to misclick on the keyboard. I had "8 excess keystrokes," but those were all due to mistyping then deleting a single mistyped character (e.g. "ee" -> "er").
Not sure how to keep the essence of the skill-based approach, without the frustration induced by typing on a mobile keyboard.
Anyway, great job and congrats again!
Just one feedback - on desktop browsers, I can see the list of answered clues below the textbox, but on the phone (Brave or Firefox on Android), I don't see that list. I am not sure if this is a feature or a bug, but it’s a feature I miss when playing on my phone. Seeing those answers gives that little “aha!” moment of satisfaction.
I also made a custom GPT - Bracket GPT [0] that helps in solving the clues when I am stuck. It doesn’t directly give the answers, but offers hints to help nudge you to the solution. It’s a fun companion when you're totally blanking.
[0] https://chatgpt.com/g/g-67e0f124cd408191943faadb3d70c6df-bra...
Countering some recurring feedback here: I find the design decisions you’ve made strike a good balance between simplicity and inviting the player in to find their way round. I’m viewing on mobile.
- I can’t easily parse a top-level clue structure but the lack of colours keeps it neat - Guess history would have been useful but omitting it keeps each guess low stakes for a first time user - The keyboard is fine, don’t let the unicode universe in for an ascii input field
# Introducing Bracket City: The Atlantic’s New Word Game
https://www.theatlantic.com/press-releases/archive/2025/04/i...
> Meet your next word-game obsession: Bracket City. Today we’re announcing that this fast-growing word puzzle - created earlier this year by an independent game designer - has a new home at The Atlantic. // In Bracket City, players solve nested layers of clues to uncover a fact about the given day in history. Each solved bracket reveals a part of the next, creating a chain reaction that eventually resolves the whole puzzle into a single sentence. Players earn city-themed ranks for success, such as “Commuter,” “Mayor,” or the coveted “Kingmaker” for a perfect puzzle. // Bracket City was created by Ben Gross and has rapidly amassed an audience of loyal players. Ben will continue to create daily puzzles with Caleb Madison, the director of games at The Atlantic.
In addition to saying congratulations, I wanted to ask, since you say you're the one writing the puzzles: how do you define a puzzle? What does it look like from your perspective? Do you literally just write all the nested clues out with brackets and everything, and then validate that it works—because that would drive me nuts, I think—or is there some tool you wrote to help?
They paid a lot less than it would have cost to rewrite and I made a lot more than if I had just kept it as a free hobby app.
My game is a bit different—it’s all about combining emojis to find the hidden emoji. Sometimes it’s wordplay, sometimes conceptual, and sometimes just plain silly.
Congrats to everyone building and sharing—loving the creativity here!
It took me longer than it should have to realize the last puzzle was made up of all my previous answers.
Spoiler: I also failed miserably at the last puzzle since I was thinking watergun and not gun so I thought it was a firefighter. I also couldn't tell that one emoji was a jail so just sort of ignored it. So that's mostly on me for having trouble understanding some of the emojis.
But more importantly we can see here that actually just getting user input with a keyboard is actually made stupidly hard on mobile, and getting a reasonable result actually requires reimplementing the virtual keyboard, which often sucks because it has a tremendous amount of complexity!
This is something that really should be fixed.
I think what might work is just if it says "Too early" and makes a list of stuff below the puzzle that you've solved too early and just put a checkmark next to it once it's ready and let you click it to enter it.
Right now it's giving negative feedback, I think it should be a positive to recognize pieces of the puzzle early, even if you still need to do the whole puzzle.
[1]: https://arstechnica.com/gaming/2022/01/whats-a-five-letter-w...
While I spend quite a bit of time here on HN, I missed the original post but started playing daily after Paris Martineau mentioned it on the Intelligent Machines (formerly This Week in Google) podcast.
I’ve been wondering how long it’d take to get picked up by some publication. It’s just too good.
I shared it with all of my friends and family when you first posted it, and we’re all hooked.
(I know you have to control the keyboard and I need to read my answers before submitting, but I’m learning how much I lean on autocorrect when putting in an answer 4x misspelled the same way)
I was kind of spoiled on today's (9th April) one because I saw a bluesky post about the same topic by depthsofwikipedia earlier today.
from now on you gotta doublecheck your Signal groups though :-)
Now get the Word of the Day ported over too!
I've been playing through the archives, and they are challenging but fun. My criticism is that sometimes the answer is very ambiguous.
For instance, (spoilers ahead), in this case
> there[something golfers' apparently [scold, with "at"]]
I know the answer is "therefore". But the inner bracket could be shout/scream/yell.
It's quite frustrating when I get docked for this, hopefully the clues can be better constructed/beta tested in the future?
Thanks again for the fun game :)
Clearly you made it as a labour of love, but if you were honest with yourself would you have made more as an uber driver, or maybe a tech employee, for the time you invested.
> whats the answer to this where you replace the bracketed sections with words?
> the [second rock from the [dial ([two-dimensional [your ["the of nowhere"] one can get you grounded] puppet projected on a wall]-based clock)]] de Milo is discovered by a Greek [guy with a [kind of stop for a [vehicle that's a palindrome ]]chfork that's not the [[[a mechanic might look under it] color for a little girl with a wolf problem] spandex halloween costume, maybe]]
Response:
Okay, let's break this down piece by piece:
["the of nowhere"] one can get you grounded] -> "middle" (middle of nowhere), but being sent to "your room" gets you grounded. Let's stick with room.
[your [room] one can get you grounded] -> room (as discussed).
[two-dimensional [room] puppet projected on a wall] -> A 2D projection on a wall is a shadow. (The 'room' clue seems less direct here, 'shadow puppet' is the key).
[dial ([shadow]-based clock)] -> A clock using shadows and a dial is a sundial.
[second rock from the [sundial]] -> The "sundial" represents the Sun. The second rock (planet) from the Sun is Venus.
[vehicle that's a palindrome ] -> racecar.
[kind of stop for a [racecar]] -> pit (pit stop).
[a mechanic might look under it] -> hood.
[[hood] color for a little girl with a wolf problem] -> Little Red Riding Hood's color is red.
[[red] spandex halloween costume, maybe] -> This describes something that is not the pitchfork. A pitchfork is not a red spandex costume. This part confirms the item described earlier.
[guy with a [pit]chfork that's not the [red spandex halloween costume, maybe]] -> Combining "pit" and "chfork" gives pitchfork. A guy with a pitchfork, especially in a Greek context finding an artifact, is likely a farmer (or peasant).
Putting it all together:
The Venus de Milo is discovered by a Greek farmer.1. You should include what you think is interesting about the large language model's ability to solve the word puzzle. It's not obvious to me as a reader.
2. You shouldn't post the step-by-step solution to a puzzle, regardless of how it was derived. It's unlikely that anyone will read through it, and if they do, they're less likely to try the puzzle themselves.
Each step (though a little confusing to parse) does make sense in context.
Getting the correct final answer tells you nothing about the reasoning. The LLM will solve the puzzle even if you only pass it the sentence "the [] de Milo is discovered by a Greek []".
The proposed intermediate answers don't make sense, and they're the actual puzzle. You must answer all the intermediates correctly, you can't skip levels.
The following ones clearly aren't correct (I think, I hadn't actually done that puzzle):
["the of nowhere"] -> middle; the model gets that right, but garbles the answer due to not correctly extracting the clue and having trailing garbage
[your [middle] one can get you grounded] -> finger; "room" as suggested by model makes no sense
[two-dimensional [finger] puppet projected on a wall] -> shadow; the model gets this right, but only by ignoring the clue completely, which means the justification is nonsense
[dial ([shadow]-based clock)] -> sun; the model says sundial, which then forces it to make an inane "a sundial represents the sun" argument at the next level.
[[red] spandex halloween costume, maybe] -> devil; the model doesn't give an answer word at all for this, but just says that "a red spandex costume is not a pitchfork"
I would note that I think the LLM had trouble because copying and pasting from the puzzle omits all the blank word underlines ("____"). I've dealt with this when copying into my own notepad for trying to work it out myself.