Treemaps are the default, and are great for finding outliers with heatmaps, but terrible in terms of explainability. So now I'm heading into architecture diagram territory, and tree-sitter is helping with extracting the names of classes, functions, variables etc. LLMs can make decent diagrams out of this.