2026-11-14
Why a game about a life uses a life table
The mechanism behind the death roll, and why a Gompertz curve with one solved parameter is a better model than a scripted ending.
Este artículo está en inglés.
Every life simulator has to answer one question before any other: when does this person die?
There are three common answers. Roll a number against a fixed distribution and stop. Script the death at a narratively satisfying age. Or ignore death until the player gets bored.
All three are wrong in the same way: they make death a property of the story rather than a property of the body and the place.
The life table
A life table is the oldest tool in demography and still the best. It answers a single question for every age: given that you are alive at age x, what is the probability you die before x+1?
For a country, this is not hypothetical. It is measured. Actuaries and health agencies publish these tables annually. They are the reason an insurance company can price a policy for a 45-year-old and not lose money.
The problem for a game is that a full national life table is a few hundred numbers, and there are eighty-three countries, two sexes, and three eras. That is a lot of table to ship, and it is a lot of table to maintain when a country’s figures are revised.
One parameter instead of a table
From about age five onward, human mortality follows a remarkably stable shape. The force of mortality rises roughly exponentially:
μ(x) = a · e^(b·x)
This is the Gompertz law, and it has held for two centuries across wildly different populations. The slope b is close to constant — mortality doubles roughly every eight years. Almost all of the variation between countries lives in the intercept a.
So instead of shipping a table, the engine ships two numbers per country: life expectancy at birth, and infant mortality. Then it solves for a numerically — sixty iterations of bisection — so that the curve, when summed into a discrete life table, reproduces the country’s stated life expectancy.
One solved parameter, and the whole mortality curve of a country falls out of it.
The first five years need separate handling
Gompertz does not fit infancy. Mortality is high at birth, falls to a minimum around age ten, then rises. A single exponential cannot do that.
So the first year uses the infant mortality rate directly. Years one through four use the under-five mortality rate to back out an average annual risk across the four years. From five onward, Gompertz takes over.
This matters because the first five years are where the largest share of the global mortality gap lives. Getting them wrong would flatten exactly the thing the game is about.
Health as a multiplier, not a second model
The four stats in the game are health, happiness, knowledge and social connection. Only one of them enters the death calculation.
Health is a multiplier on the hazard, running from 0.55 at health 100 to 3.2 at health 0. A perfectly healthy person in a given country dies at 55% of that country’s baseline rate. A person in terrible health dies at 3.2 times it.
Knowledge never enters. This is deliberate and it is enforced structurally: the death function’s signature does not accept a knowledge value at all. A 40-year-old with knowledge 95 and health 20 dies at the same rate as a 40-year-old with knowledge 20 and health 20.
That single structural decision is what separates this from a game where studying is a strategy for not dying.
Why a roll per year
The alternative — compute a single death age up front, then narrate toward it — is cheaper and much easier to make dramatic. It is also dishonest in a specific way: it makes the death feel earned. If the player dies at 63, a pre-computed model invites them to look back for the reason. There is no reason. There is a roll.
Rolling once per year means the same seed always produces the same life — the random stream is deterministic — but the player experiences it as genuine uncertainty, because at every step it is.
What it cannot do
It cannot tell you anything about you. It models a population and draws an individual from it. If you want to know your own life expectancy, this is the wrong instrument, and no web page is the right one.
What it can do is make a number you have read a hundred times into something you sit inside for half an hour. That is a different kind of knowing.