Add a graphify knowledge graph of ICSharpCode.Decompiler

[christophwille]

Adds a graphify knowledge graph of ICSharpCode.Decompiler/, committed under graphify-out/.

  graph.html            - interactive graph, open in browser
  GRAPH_REPORT.md       - audit report
  graph.json            - raw graph data

Built on ICSharpCode.Decompiler/ (583 files), narrowed from the 2,145-file repo. 9,519 nodes / 27,828 edges / 283 communities. Extraction was 9,615 AST nodes (free, deterministic) plus 39 semantic nodes from 2 docs and 1 image.

Two caveats worth stating plainly:

• Graph health warning: 1,051 dangling-endpoint edges (references to types outside the scanned directory - expected when scoping to one project), and 4,324 edges collapsed into 27,828 unique undirected pairs. The collapse is mostly benign multiplicity - e.g. DecompilerSettings has 121 separate bool field references that fold into one edge - but it means edge counts understate raw reference density.
• Token accounting: the harness reported combined subagent totals (68,911) without an input/output split, so cost.json records it all as input.

Benchmark: 7.1x fewer tokens per query vs. naive corpus reads (~634,600 -> ~89,918).

God Nodes

#	Node	Edges
1	IType	631
2	ILInstruction	600
3	System	315
4	ICSharpCode.Decompiler.TypeSystem	252
5	AstNode	213
6	ILVisitor	207
7	IMethod	193
8	CSharpOutputVisitor	183
9	ExpressionBuilder	180
10	ICSharpCode.Decompiler.CSharp.Syntax	171

Surprising Connections

• ForeachAnnotation --references--> ILInstruction - CSharp/Annotations.cs -> IL/Instructions.cs
• ImplicitReturnAnnotation --references--> Leave - CSharp/Annotations.cs -> IL/Instructions.cs
• QueryGroupClauseAnnotation --references--> ILFunction - CSharp/Annotations.cs -> IL/Instructions.cs
• QueryJoinClauseAnnotation --references--> ILFunction - CSharp/Annotations.cs -> IL/Instructions.cs
• ILVariableResolveResult --references--> ILVariable - CSharp/Annotations.cs -> IL/ILVariable.cs

Suggested Questions

• Why does IType connect 80+ communities spanning the IL layer, the C# syntax layer, and the type system?
• What is the relationship between the NuGet icon's visual metaphor and its package-branding role? (AMBIGUOUS edge, low confidence)

The Annotations.cs cluster is the standout: it is the single file where the C# AST layer holds direct references back into IL instructions - the bridge that lets decompiled syntax nodes carry their originating IL. Every surprising connection the graph found lands there.

Notes on what is and is not committed

graph.json is 18 MB and is committed in full, so the graph is queryable from a clean clone without re-running extraction. The AST/semantic caches are excluded via .gitignore - they key off absolute paths on the machine that produced them and are not portable.

🤖 Generated with Claude Code

[christophwille]

Claude added one Q: ▎ The most interesting question this graph can answer: How does CSharp/Annotations.cs bridge the IL layer and the C# syntax layer, and what does that coupling let the decompiler do? Want me to trace it?