We gave an LLM a structural graph of a codebase before exploring. It used 54% MORE context than without one. Paper + explanation inside [R]

In three linesControlled study on TypeScript codebase (25 sections, 3,250 files): LLM (Kimi K2.6) equipped with structural graph (Blueprint: Universal Ctags + ast-grep + BM25) consumed 54% more input tokens (63,541 vs 41,327) but explored deeper (6 turns vs 5). Graph costs ~6,500 tokens and increases model's navigational confidence.

Summary generated by Claude — human-verified