Research · Research overview
Evernormal: Exploring Intelligence Through a Financial World Model
A closer look at a simulated household economy, the data behind it, and the questions it opens about AI decision-making.
What can real-world data tell us about intelligence? Evernormal approaches that question through a financial world model: a simulated environment in which household circumstances, daily decisions and their consequences can be examined together.
Grounding a simulated world
Evernormal’s website describes a population assembled from U.S. survey microdata, with income, spending and balance-sheet information feeding a daily household simulation. Its public city viewer makes simulated household finances and decision journals visible. These are synthetic outputs, not records of identifiable people.
From circumstances to decisions
The technical report separates the environment from the decision-maker. It describes rule-based and language-model-driven human simulation, alongside a delegated financial-agent application that is partially built. This distinction makes the research question concrete: what changes when a different decision-maker faces the same circumstances?
Reading the evidence
The project publishes validation comparisons, residuals and documented assumptions. Those reports are the project’s own evidence and should be read with their stated scope; a simulation result does not establish how an AI system will perform in real households. The report and viewer below offer a starting point for exploring the model and its limits.