You open Claude Code. It reads your vault. It answers with a fact that stopped being true 30 days ago. You wrote that note once and never opened it again. Neither did the agent, until it needed it. Meanwhile, developers on Reddit are running vaults that cut session context by 67% and score 100% recall against a golden set. One of them runs a loop every two weeks that catches drift before he sees it. The difference isn't talent. It's a loop that runs while he sleeps. This book gives you that loop Saturday you build the constitution and the fact schema. Sunday your first rot report names the pages your vault is already wrong about. By next weekend the loop runs unattended and sweeps your busiest pages first. By day ninety a regression suite proves your second brain still answers accurately. This isn't another "intro to Obsidian" book. Every artifact ships in a free companion repository: the constitution, the schema, the sweep, the hooks and the scheduled loop, at the paths the book uses, with one dry-run command that shows what it would change in the vault you already have before it changes anything. Here's what you'll build: 1. A CLAUDE.md vault constitution the agent cannot write past. 2. A page schema with three fact states: current, superseded, disputed. 3. rot-report.md, naming the stale and contradicted pages in your vault tonight. 4. A review gate: the agent stages, you approve, nothing lands unreviewed. 5. INDEX.md, agent-maintained, plus your own measured drop in session context. 6. The dreaming loop, scheduled and unattended, sweeping your busiest pages first. 7. A client package priced at $2,200 setup plus $1,500 a month. Every week another 92-page "second brain" short read lands in the Kindle Store. This one is 180+ pages of schemas, sweeps and scripts you run against the vault you already have. Scroll up and grab your copy.