Writing forty words that have to carry a decision
Most of the writing in a card-driven game is invisible as writing. Nobody finishes a run and says the prose was good. What they say is that a decision felt heavy, or that they thought about a card afterwards, or — the failure case — that they stopped reading the cards about twenty minutes in and started pattern-matching the buttons.
That last one is the thing all of this is trying to prevent. An encounter card has a speaker line, three or four lines of body, and two option labels. Forty to sixty words total, on a phone, competing with a bus window. Within that budget it has to establish a situation, imply what each choice will cost, and then get out of the way.
The register, and why it never winks
The whole deck is written in one voice: dry, plain, slightly grim, occasionally deadpan. It never nudges the player, never jokes about the genre, and never acknowledges that it is a game.
That is a decision with teeth, because the winking register is so much easier to write. A card about a preacher asking for a tithe practically writes its own knowing aside. The reason not to is that a wink costs you the only thing this kind of prose has going for it. Forty words cannot build a world through description; there is no room. What they can do is behave as though the world is already there and the reader arrived late — and the instant the text winks, it admits there is no world, only a designer.
The register also does mechanical work. When the tone is flat, the grim lines land harder without being underlined. A landmark blurb that says only it's a rock is funnier and bleaker than any joke I could have written about it. A hot spring described as something the children think is a miracle, followed by four words saying it isn't, does more characterisation of the narrator than a paragraph would.
The rule I settled on: the text may be bleak, and it may be dry, but it may never be pleased with itself.
The speaker line is the cheapest worldbuilding available
Every card names who or what is talking before the body text. Sometimes it is a person — a wagon master, a travelling preacher, a family on the roadside. More often it is not: the sky, the trail, the camp after dark, the near ox, a line of smoke.
Attributing a card to the sky costs two words and does an enormous amount. It tells you nobody is going to explain this to you. It sets a scale — the thing addressing you is weather, and weather has no opinion about your party. And it makes the human speakers land harder by contrast, because when an actual person shows up and says something, they are interrupting a landscape that has been talking at you for an hour.
It is also a discipline for me. Writing a card starts with deciding whose card it is, and that decision usually reveals whether the card is any good. If the honest speaker is "the game," the card is a mechanic wearing a costume and should be cut.
Option labels are the hardest forty characters in the project
A label has to be short enough to sit on a button, specific enough that you know what you are choosing, and honest about the price. Those three pull against each other constantly.
The failure mode I keep catching is a label that promises a transaction the game does not perform. An option called pay the ferryman implies coins leaving your purse. If the underlying card does not actually charge coins, the label has lied, and worse, it has lied in a way that makes the option strictly better than its alternative — you get the benefit of paying with none of the payment. A deck audit catches these, but the real fix is upstream: every verb in a label is a promise, and the promise has to be kept in the numbers.
There is a subtler version. A repair option might read mend the wheel, and if you have the right spare it genuinely mends it. If you have nothing — no part, no tools, nobody who knows a forge from a fireplace — the option quietly resolves as the other one: bind it and hope. That is good design, because it means no choice is ever a dead end. But the button still says mend, and a player who does not understand why it did not mend has been misled by their own interface. The fix is not to remove the fallback; it is to say so, in the label or the outcome, in words.
Writing around a hole the machine will fill
Several cards involve a specific member of your party, chosen at runtime — the weakest person, or a random one. The text is written with a placeholder that gets replaced with a name when the card is dealt.
This is a genuinely awkward thing to write well, and the awkwardness is easy to underestimate. The sentence has to work with a name that could be one syllable or four, could belong to a man, a woman, or a child, and might be someone the player has grown attached to or someone they barely remember hiring. It cannot lean on a pronoun that might be wrong. It cannot be built around a rhythm that a long name destroys.
What works is to give the placeholder the subject position and keep the sentence short and factual around it — the way you would actually be told bad news by someone who did not want to say it twice. One card runs, roughly: this person has been saying their stomach hurts for two days; this morning they stopped saying it, which is worse. The name goes in the first clause, the sentence gets out of its own way, and the second half does all the work regardless of who was named.
What does not work is any construction that needs to know something about the person. The moment the text implies age, or relationship, or how long they have been with you, it will be wrong for some fraction of runs — and that fraction will notice.
Placeholder text is a debt with interest
There is no placeholder text anywhere in the deck. Not one TODO, not one lorem line, not one card whose body is a description of the card.
This is a rule I adopted after violating it, and the reason is that placeholder text in a content-driven game does not read as unfinished. It reads as bad. A player encountering a card whose body says "storm event, medium severity" does not think ah, still in development; they think this game is cheap, and they are right to, because it is the same experience either way. And the debt compounds: once there are twenty placeholders, writing them all becomes a project, and projects get deferred. Twenty cards written at the moment each was designed is an afternoon spread over months. Twenty cards written at the end is a week nobody has.
So a card does not exist until its text exists. It is a slower way to build a deck and it is the only way I have found that produces a deck where every card is worth reading.
The near-duplicate problem
One line appears twice in the game — a remark about a river being deeper than it looks — once on an encounter and once as a landmark's description. I found it in a writing pass and left it.
That is a judgement call and I still think about it. In prose, a repeated line is either a deliberate refrain or a mistake, and there is no third option. Here it reads as a refrain: the same practical warning, offered twice, by a country that only knows one thing about water. But it is close enough to the line that a reader could reasonably read it as sloppiness, and the difference between refrain and repetition lives entirely in whether the reader trusts you.
Which is the honest summary of writing at this length. You do not have the space to earn trust with craft on any single card. You earn it across fifty of them, by never once being sloppy where it shows — and then you get to spend a little of it on a repeated line about a river.