Hiring people instead of buying gear
The party in a trail game is the most expensive purchase the player ever makes and the one they understand least at the moment they make it. You are at the jumping-off town, you have a purse, and you are choosing between four people out of six with nothing but a job title and two numbers to go on. Everything downstream depends on it and none of it is legible yet.
Designing that roster turned out to be less about the characters and more about answering one question honestly: what are you actually buying when you hire somebody?
Two stats, and what they are allowed to mean
The roster runs on two numbers per person, and holding it to two was a deliberate constraint. Every additional stat multiplies the comparison work the player has to do at the exact moment they have the least context.
The first is a body stat, and its meaning is specific: it is how many hits this person can absorb before the trail kills them. Not strength, not toughness in the abstract — capacity. It maps to a small integer, roughly two to six, and that integer is the whole of it.
The second is a reading stat: how well this person reads a situation. It drives the odds on every card that asks you to send somebody to do something delicate — judge a ford, sort spoiled stores from good, find the safe line down a grade. Those odds are clamped at both ends, so your sharpest person is never a certainty and your dullest is never hopeless. That clamp is doing quiet work: it keeps a person-card a decision about who you can spare rather than a lookup of who has the highest number.
Two stats is not many. It is enough because the real differentiation lives elsewhere.
The perk is the character
Each person carries one standing effect that applies for the whole run. A widow whose presence makes the party take sorrow and sickness about fifteen percent more softly. A preacher who starts the party in better spirits and lifts them again on any layover of a couple of days. A wheelwright who makes the wagon take damage a little less badly. A cook who cuts food consumption by around a fifth. A schoolmistress under whom good days count for more.
Those are the characters. The two stats tell you whether somebody survives; the perk tells you what the run feels like with them in it. Take the preacher and you are playing a game where stopping is rewarded, so you stop, so you arrive late, so winter becomes your problem instead of starvation. Take the cook and the calendar loosens. The perk changes the strategy, and the strategy changes which cards are hard.
There is a trap I fell into and want to name: a perk that does nothing. One character carries an attribute labelled as reading the ground, with a value of one — and the code that applies perks explicitly skips it, because that character's ability is already expressed through their high reading stat. The value is inert. It is harmless in play and genuinely dangerous in maintenance, because it looks like a knob. Somebody — probably me, in a year — would have tuned it, watched nothing happen, and gone looking for a bug in the wrong file. Anything that looks like a dial and isn't should be either removed or documented in the data itself.
What hiring a tradesperson actually buys
This is the part I got wrong first, and the correction reshaped the economy.
The original idea was that a tradesperson arrives carrying their tools: hire the smith and you get his toolset, which is worth more than his fee, and the bargain is obvious. It is a clean pitch. It is also a mess, because it makes hiring strictly better than shopping — you are buying the tool at a discount and getting a person attached for free — and it means the shops that sell those same tools become traps for anybody who has not worked it out.
What ships instead: a tradesperson gives you a standing discount on their entire trade line, everywhere, for the rest of the run — about a quarter off — plus their perk and their body. They hand you nothing at the moment of hiring.
That version is better on every axis. It preserves the shops, because you still have to buy the toolset; it just costs less now. It makes the hire a bet on the future rather than an immediate swap, which is a more interesting decision and a riskier one. It scales with how long the run has left, so hiring the same specialist early and late are genuinely different propositions. And it gives the pricing a rule I can state: a premium hire should cost a little less than the single tool they most obviate. The player who does the arithmetic discovers a bargain; the player who doesn't is not punished for it.
The problems the roster has
An honest register of what is still wrong, because the register is more useful than a tidy story.
A free character is out-competing a paid one. The hunter is the only person on the roster with top marks in both stats, is free in the starting pool, and carries a perk that multiplies food gained from foraging. On his own card the food reward is already scaled by his reading stat, and then multiplied again by the perk — a stat-driven reward getting a second multiplier stacked on it. Meanwhile the cook, who is the paid food specialist, delivers a smaller food swing for real money. A free character beating the paid specialist at the paid specialist's own job is not a difficulty problem; it is a pricing problem, and it makes the cook look like a mistake to anybody who has played twice.
Two characters do the same job. The smith and the wheelwright both reduce wagon damage, at different strengths. Take both and they stack, down to a floor. Two of ten characters occupying one design slot is a lot of roster to spend on redundancy, and the second one you hire feels like a smaller purchase than the first, which is a bad feeling to sell somebody.
Two more characters do the same job, later. The midwife and the physician both improve tending the sick, both work the same trade, both unlock in the back half. The physician is better on both stats and the stronger healer; he costs a little more and appears a few hundred miles later. That is a real trade only if the mileage gap genuinely bites, and I am not yet convinced it does.
Everybody paid is fragile. Half the roster sits at the lowest survivable body capacity, and all of the paid hires are in that half. You pay for perks and trade discounts; you never pay for toughness. That is defensible as a theme — the people worth hiring are specialists, not oxen — but it means a late-game party is a collection of valuable, breakable people, and the game should probably say so out loud rather than letting the player discover it at a river.
What the simulation says about all this
The balance harness can measure some of these and not others.
It can tell me that hoarding is unwinnable and that over-caution starves, which indirectly prices the cook's food perk — anything that loosens the calendar is worth more than it looks, because the calendar is what kills. It can tell me which cards are effectively auto-picks, which flags the person-cards whose alternatives are dominated.
It cannot tell me whether the free hunter feels overpowered, because the harness's parties are held constant on purpose — it varies decisions, not drafts, so that a policy comparison measures in-run skill rather than provisioning luck. Sweeping party composition is a separate experiment with a different design, and it is the next thing to build.
Which is the ordinary condition of this work. You measure what the instrument measures, you write down what it can't, and you resist the urge to let the measured questions crowd out the important ones just because they answer faster.