Utterly Predictable

The science of why humans do exactly what you'd expect

A behavioral prediction engine that takes crisis scenarios and outputs group-by-group predictions of human response — grounded in peer-reviewed behavioral science, not punditry. We predict what people will do, not what they say they'll do. And we show you the gap between the two.

Powered by: Reference Class Forecasting • Moral Foundations Theory • Social Identity Theory • Terror Management Theory • Prospect Theory • Superforecasting

Choose a Scenario

Select a crisis scenario to see group-by-group behavioral predictions. Each analysis uses a 5-layer pipeline: event classification, historical reference class, group segmentation, temporal dynamics, and behavioral output.

What 60 Years of Disaster Research Actually Shows

These findings recur across every crisis we studied. They contradict most popular assumptions.

Mass Panic Is a Myth

The Disaster Research Center (est. 1963) has 60 years of data. Populations are overwhelmingly prosocial in crises. "Elite panic" — authorities deploying military, protecting property, restricting movement — is the documented phenomenon.

Political Identity Outpredicts Everything

In the modern West, political affiliation is the #1 predictor of crisis behavior — beating age, education, income, and health status. COVID mortality diverged along party lines more than any medical variable.

The Say/Do Gap Is Universal

Preference falsification (Kuran) means public behavior never equals private belief. The leading indicator of cascades: widening delta between anonymous behavior (search data) and public behavior (social media, polls).

Historical Parallels Beat Theory

Tetlock's outside view: don't model from first principles. Find the closest historical reference class, use the base rate as anchor, adjust for novel elements. Algorithms beat experts because they're noise-free.

Humans Adapt to Almost Anything

Hedonic adaptation: 3-6 months for most constant stimuli. Exceptions that never adapt: chronic food insecurity, chronic unpredictability, and chronic lack of control.

25% Tips It, 3.5% Topples Governments

Centola: visible behavior locks in when 25% adopt. Chenoweth: no government has survived a sustained nonviolent campaign with 3.5% active participation. Below threshold: nothing. At threshold: sudden phase transition.

Opinion Radicalization Is Not Action

McCauley's Two Pyramids: millions think radical, very few act radical. Different mechanisms. Movement from opinion to action is driven by network closure and status deficit, not ideological intensity.

Framing Determines Response More Than Facts

Kahan, Feinberg and Willer: the same threat produces opposite responses depending on whether it's framed through the audience's values or against them. Identity-protective cognition overrides data — until material cost exceeds social cost.

The Accuracy Ceiling Is ~60-70%

Kahneman's "Noise" research: predict categories and directional probabilities, not exact magnitudes. Beyond 3-5 years, accuracy approaches chance. We show confidence levels, not certainties.

Get Predictions for New Scenarios