RAG Interview Questions #24 - The Resolution Limit Trap

You’re in a Senior ML Engineer interview at Microsoft and the interviewer asks:

“You built GraphRAG with Leiden for community detection. It’s deterministic, gives clean hierarchies, everyone’s happy. Now your retrieval quality is degrading on certain queries. When does maximizing modularity actively hurt you?”

Don’t say: “Leiden maximizes modularity to find well-connected communities, and it’s better than Louvain because it guarantees connected partitions.”

Here’s the reality:

Modularity is not a measure of “correct” communities. It’s a measure of density. Those are not the same thing, and in production, the gap between them is where your retrieval breaks.

Modularity rewards one thing: nodes more densely connected inside a community than you’d expect by chance. Maximize it hard enough and you inherit its failure modes: