Genetic drift, and why small populations lose variety

Not every change in a population is an adaptation. A great deal of it is luck: variants becoming common or vanishing entirely for no reason except who happened to reproduce. In a large population the luck averages out. In a small one it does the driving, and it is often stronger than selection.

Why chance is not noise

Imagine a population where half the individuals carry one version of a gene and half carry another, and the two versions do nothing different. Each generation, the parents of the next generation are a sample of this one. Samples are not perfect copies of what they are drawn from.

So the proportion shifts a little, at random. And here is the part that surprises people: that shift does not tend to correct itself. The next generation samples from the new proportion, not the original one. There is no restoring force. The frequency wanders, and because it wanders without a pull toward the middle, it eventually hits one end or the other.

When it hits zero, that variant is gone from the population permanently, unless mutation reintroduces it. When it hits one hundred per cent, every alternative is gone. Either way, variety has been lost, and nothing about the organisms caused it.

Drift against selection

Both processes run at the same time, and which one dominates depends on two quantities: how strong the selective advantage is, and how big the population is.

A variant with a large advantage in a large population will spread reliably, because the advantage overwhelms the sampling noise every generation. The same variant with a very small advantage in a small population may well disappear, because the noise is larger than the signal.

This has a consequence that sounds wrong the first time you meet it: in small populations, mildly harmful variants can spread and become universal, and mildly beneficial ones can be lost. Selection does not stop working. It stops being the loudest thing in the room.

Motoo Kimura's neutral theory pushed this further, arguing that the large majority of genetic change at the molecular level is not adaptive at all, but neutral variants drifting to fixation. That was controversial and is now broadly accepted for much of the genome. It reframes the default assumption: a difference between two populations needs evidence before it is called an adaptation, because drift is the null explanation and it is always running.

Bottlenecks and founders

Two situations concentrate drift enough to reshape a population permanently.

A bottleneck is a sharp reduction in numbers. Whatever survives carries only a sample of the variety that existed, and the sample is not chosen for merit, because a catastrophe is generally indifferent to fitness. When the population recovers, it recovers from that sample. Numbers return. Variety does not.

A founder effect is the same arithmetic from the other direction: a small group establishes somewhere new, carrying an accidental subset of the original variety, and everything descended from them inherits that accident. Whatever was rare among the founders may be universal in their descendants purely because of who made the trip.

In both cases the population's genetic composition has been set by an event that had nothing to do with what worked. This is a large part of why conservation biology cares about numbers well above the point of immediate survival: a species can be numerically recovered and genetically impoverished at the same time, and the second condition is not visible from a headcount.

Where this shows up in a world you can watch

This world runs small populations by biological standards, and it oscillates, which means it passes through bottlenecks routinely. Drift is not a subtle background effect here. It is a leading character.

Every organism carries nineteen genes and a couple of thousand network weights, all inherited with mutation, and lineages are tracked individually with a stated rule for when a group has diverged enough to count as its own. So when a lineage's typical gene values shift, there are two candidate explanations available, and the honest position is that separating them requires more than observing the shift.

That is not a limitation peculiar to this project. It is the same problem field biology has, and it is the reason a controlled comparison matters more than a striking observation. What this world adds is that the comparison is actually available: the same run can be repeated exactly from its seed, and a control condition can be run alongside it, which is the point of determinism and reproducibility.