Schelling Segregation Simulator

What you just saw

Agents with mild preferences about their neighbors produce extreme segregation. Even at a similarity threshold of 0.3 — wanting only 30% of your neighbors to be like you — the grid self-sorts into clean neighborhoods within a few dozen steps. The individual rule is tolerant; the aggregate outcome is not.

The science behind it

Thomas Schelling introduced this model in 1971 to show that segregation can emerge even when no individual prefers it. The model is agent-based: each cell is a person with a type and a rule. At every step, unhappy agents (those with too few same-type neighbors) relocate to random empty cells. The rule is local; the segregation index — average same-type-neighbor fraction — is the emergent measurable. The model is the conceptual foundation of computational social science.

Try these experiments

1. 1. Tolerance floor

   Settings: similarity_threshold = 0.25

   What to look for: The grid stays mostly mixed. A modest same-type preference does not force segregation if it is below the tipping threshold.

2. 2. Tipping point

   Settings: similarity_threshold = 0.35

   What to look for: Clear neighborhoods form within ~50 steps. The segregation index climbs from ~0.5 to ~0.8.

3. 3. Sharp segregation

   Settings: similarity_threshold = 0.6

   What to look for: Complete neighborhood separation. Agents cluster quickly and segregation index approaches 1.0.

4. 4. Paralysis

   Settings: similarity_threshold = 0.9

   What to look for: Almost no one is ever satisfied. The system thrashes — many moves per step and high unhappiness.

from scirouter import SciRouter

client = SciRouter(api_key="sk-sci-...")

result = client.simulation.run(
    model="schelling",
    params={
        "grid_size": 500,
        "density": 0.9,
        "similarity_threshold": 0.35,
    },
    steps=500,
    seed=42,
)

print(result.metrics["segregation_index"])