SEIR Epidemic Simulator

What you just saw

A population moves around a shared space. Infected agents transmit to susceptible neighbors within a contact radius, and the familiar epidemic curve rises, peaks, and falls. The shape of the curve — when it peaks, how tall it gets, how much of the population ever gets exposed — is set almost entirely by three numbers: transmission rate, incubation period, and recovery period.

The science behind it

The SEIR model divides a population into four compartments — Susceptible, Exposed, Infected, Recovered — and tracks the flow between them. It is the workhorse of infectious disease modeling and was parameterized extensively with real COVID-19 data during 2020-2022. The agent-based implementation here gives you the same dynamics but with spatial structure: transmission depends on proximity, not just mixing, so you can see wavefronts of infection sweep the grid.

Try these experiments

1. 1. Slow burn

   Settings: transmission_rate = 0.1

   What to look for: Long, shallow curve. Endemic equilibrium possible.

2. 2. Classic wave

   Settings: transmission_rate = 0.3

   What to look for: Clean wave shape with a single peak around day 60-90.

3. 3. Explosive

   Settings: transmission_rate = 0.6

   What to look for: The wave finishes almost immediately. Most of the population is infected in ~30 days.

from scirouter import SciRouter
client = SciRouter(api_key="sk-sci-...")
result = client.simulation.run(
    model="seir",
    params={"population": 5_000_000, "transmission_rate": 0.3,
            "incubation_period": 5, "recovery_period": 14,
            "mortality_rate": 0.015},
    steps=365,
    seed=42,
)
print(result.metrics["peak_infections"])