worked: add opencode TypeScript corpus (406 files, 12.6x reduction, honest review)

[sdeonvacation]

What this adds

A new worked example in worked/opencode/ — a real production run on the opencode codebase (packages/opencode/src/, 406 TypeScript/TSX files).

Why opencode

• Real production code, not synthetic. A TypeScript AI coding assistant using Effect-ts, Solid.js, SQLite, Vercel AI SDK.
• Different failure modes than the existing examples (Python/karpathy). Shows how graphify behaves on a functional TypeScript codebase with short common method names.
• Honest review — scores 4.3/10 overall, calls out 3 specific bugs with repro steps.

Results

Metric	Value
Files	406 TypeScript/TSX/JS
Nodes / Edges	1,772 / 2,958
Token reduction	12.6x
Extraction cost	0 (AST-only)
Real communities	38 of 213 (175 single-file isolates)

Bugs found and documented in review.md

1. Generic method names dominate god nodes — push(), get(), set() with 40–72 edges each. Useless for TypeScript codebases with functional patterns. All 10 god nodes are JavaScript built-in method names, none are architectural concepts.

2. INFERRED edges appear in AST-only graphify update run — 34% of 2,958 edges are INFERRED (avg confidence 0.8) after an AST-only update. Expected: 0% INFERRED. At least one verified false positive: killTree() --calls--> sleep().

3. TypeScript barrel file re-exports not resolved — Modules using index.ts re-exports appear as isolated single-file communities. 175 of 213 communities are single-file isolates.

Checklist

• Real run (not simulated)
• Honest scores with specific findings
• Bugs documented with repro steps
• README.md, GRAPH_REPORT.md, graph.json, review.md all present