Three rules for making a game hard without letting it cheat
Difficulty is the easiest thing in game design to fake. You want a game to feel hard, so you make the numbers meaner, and the result feels hard in exactly the way a rigged carnival game feels hard — which is to say, it feels like being cheated, and players can tell the difference within about ten minutes even when they cannot articulate what is wrong.
Getting it right turned out to need three rules, written on the wall, outranking every other decision in the project. They came out of arguing about a single design document for a frontier town and discovering that most of the arguments were really about one of these three.
Rule one: nothing may counteract itself
No mechanic may eat what it just gave you.
This sounds obvious and is violated constantly, usually by systems that were designed separately and met later. The classic form: you introduce a mechanic that lets a player buy information about what is coming — and you charge for it in the same resource the information is about. Now the player who is short on days must spend days to be told they are short on days. The mechanic hands out a benefit with one hand and confiscates it with the other, and the net effect on a struggling player is negative.
Once you name the rule you start seeing it everywhere. A rest mechanic that restores health but advances a disease clock. A scouting ability that reveals hazards and makes noise that attracts them. A repair option that fixes the wagon by consuming the parts you would have needed for the next break. Each of these is defensible in isolation — "it's a trade-off!" — but a trade-off is only honest when the two sides come from different pockets. If both sides draw on the same pocket, you have not designed a decision, you have designed an illusion of one with a hidden negative expected value.
The practical consequence in this game is that information is free. A shopkeeper who tells you the pass is closing early does not charge you for it, does not cost you a day, and does not require a skill check. It is the one thing in the whole design that reduces uncertainty without consuming a resource, and it is free precisely because charging for it would violate rule one.
Rule two: hard must be provably solvable
If someone picks the hardest setting, it should be genuinely hard — and still winnable to the end.
The word doing the work there is provably. It is very easy to write a hard mode by turning up a multiplier and shipping it on the assumption that somebody skilled will manage. Sometimes that assumption is wrong and nobody finds out for months, because the players who would have found out quit at hour two and told you nothing.
So difficulty is not allowed to live in rigged blows. It lives in three places, all of them visible to the player before the run starts: how much money you begin with, how long the good season lasts, and consequently who you could afford to bring. The trail itself hits every starting position identically. A hard run is hard because you arrive at the mountains with less, not because the mountains got meaner when you weren't looking.
That constraint is what makes proof possible. If the trail is constant, you can put a machine on it — a harness that plays the real game engine tens of thousands of times per configuration with a range of bot strategies — and measure whether the hardest starting position is winnable and by how much skill. That harness is the other half of this rule. Without it, "hard but fair" is a hope. With it, it is a number you can read off a table, and the number is allowed to be wrong, which is the whole point of measuring.
The measurement has already overturned received wisdom twice. One starting position had a reputation, carried over from an earlier version of the rules, as effectively unwinnable — a four-percent curiosity for masochists. In the current rules it is not merely winnable; for an unskilled player it is the easiest of the four, because the extra person it lets you bring and the larger larder it starts with more than pay for its other disadvantages. Nobody would have guessed that. Nobody had to.
Rule three: the trail sets a tempo — it does not hunt you
This one is the newest, and it is the one that changed the most cards.
The encounters shouldn't intentionally kill anyone. They should set a pace of play — so that if you don't keep up, the conditions stop being good all the time.
Nothing in the game reaches out and kills you. You die of arithmetic: a hundred small subtractions, accumulated while you were slow. The encounters are a metronome. Fall behind the beat and the conditions stop being kind — not because the game noticed you were struggling and pressed harder, but because winter arrives on the date winter was always going to arrive.
The distinction matters enormously to how a loss feels. A game that hunts you produces the sensation of being singled out; every bad card feels like the designer's thumb on the scale. A game that keeps a schedule produces something closer to how the historical journey actually failed people. The river did not want you dead. It was just a river, and you were late.
The hard constraint that follows is simple to state and unforgiving to enforce: no single card, including the worst weather, may take enough to end a run that started the day healthy. If it can, it is not a tempo — it is a trap. Every card in the deck is checked against that, with numbers, and the ones that came close were rewritten rather than argued for.
This is also why the game's storms have dates. A storm is not a random event that lands on you; it is a weather state with a name and a day, announced in advance by somebody who had reason to know. It turns a slow decline into a race to a specific place by a specific date, which is a far better shape for a decision, and it puts the outcome squarely on the player's planning rather than on the shuffle.
Where these rules fight each other
They do fight, and the fights are the interesting part.
Rule one says the player should never pay to learn something. Rule two says hard mode must actually be hard. Put those together and there is real tension: free information makes the game easier, and the easiest way to claw back difficulty is to charge for the information. That is exactly the move rule one forbids. The resolution is that difficulty gets restored somewhere honest — a shorter season, a thinner purse — rather than by re-introducing a toll on knowledge.
Rule three says nothing may end a healthy run in one blow, which pushes every individual card toward mildness. Rule two says the hardest setting must genuinely threaten. The resolution there is that danger is cumulative by design: the game is allowed to kill you over four hundred miles and not over one card. That is a harder game to tune, because the failure mode is invisible in any single card and only shows up in aggregate — which is, again, why the simulation exists.
The test I actually use
When a mechanic feels wrong and I cannot say why, I ask three questions in order. Does this mechanic take back what it just gave? Could a skilled player still win with this at its worst setting, and can I show that rather than assert it? Could this, on its own, end a run that was in good shape this morning?
Almost every bad idea I have had in two years of this fails one of those three. A surprising number fail more than one. And the ones that pass all three have, so far, been worth building — which is about as much as you can ask of a rule.