Homophily under model substitution: working interface
Working interface

Why

Replication under model substitution tests whether an apparent social mechanism is robust to the language model used to instantiate it.

Approach

A force-directed graph makes community formation visible over rounds, paired with a diversity score and bootstrap analysis.

Results

The live dashboard layers the simulation, comparison, and methodology for readers with different levels of technical depth.

Things I got wrong

The first visual explanations assumed too much graph-theory fluency. Showing the network settle over time made the mechanism clearer than adding more annotation.

Future work

Complete the written replication note and extend the comparison across model families.