Agent-based modelling and spatial economics

The models from Dan Olner's 2013 PhD, running live in the browser. Each one is the thesis' own experiment, rebuilt from the original Java and re-verified against the published figures.

Every model page has notes on what to watch, the economics behind it, how it is actually built — and, where there is one, what the rebuild found that the thesis did not say. Those disagreements are the point rather than an embarrassment: one figure turned out to be plotting a different variable than its label claims, and one result still has not reproduced.

If you want the argument before the models, start with how these models are built — and what that costs: why an agent's budget is a day rather than money, which kind of decision quietly breaks, and why showing something is the start of an explanation rather than the end.

The rebuild was done with Claude Code, Anthropic's AI coding agent, working from the thesis, its LaTeX sources and the 2013 Java. Claude Code read all three, rebuilt the models first in Java to check them against the published figures and then in JavaScript for this site, and wrote the code, the tests and the notes on each model page. Dan directed the work, made the design decisions and checked the results. Two of the conversations are published: the main build, from a first reading of the thesis to this site going live; and green blob, red blob, which decodes a 2011 video from the years of building the framework and rebuilds it as H1, the curiosity at the end of the catalogue.