Keep raw inbox captures out of AI retrieval scope
Index only processed permanent notes in AI-searchable systems while keeping unprocessed inbox captures outside retrieval scope, because AI systems cannot distinguish epistemic status and will retrieve raw captures with equal confidence to verified knowledge.
Why This Is a Rule
AI retrieval systems have no concept of epistemic status. A half-formed thought captured at 2 AM, a verified conclusion refined over three weeks, and a quote you haven't fact-checked yet all look the same to an embedding model. When you index everything in your knowledge base — inbox, drafts, permanent notes — AI retrieves raw captures with the same confidence as verified knowledge. Your future self, or anyone querying your system, cannot tell which is which.
This rule creates an architectural boundary: only processed, permanent notes enter the AI-searchable index. Inbox captures, fleeting thoughts, and unverified material stay in a separate scope that AI cannot access. Processing — reviewing, verifying, rewriting in your own words, connecting to existing notes — is the quality gate that determines what earns retrieval trust.
When This Fires
- Setting up semantic search or AI copilot features in your knowledge tool (Obsidian, Notion, custom RAG)
- Deciding which folders or tags to include in your embedding index
- Noticing AI retrieving half-baked thoughts as if they were established knowledge
- Building any system where AI draws from your personal notes to generate responses
Common Failure Mode
Indexing your entire vault "for completeness." You get better recall (more results) at the cost of catastrophic precision (most results are garbage). A quick inbox capture like "maybe X causes Y??" gets retrieved alongside your refined analysis of the same topic, and the AI synthesizes them as if they carry equal weight. The exploratory doubt and the verified conclusion merge into confidently wrong output.
The Protocol
Create an explicit boundary in your knowledge system: an "indexed" zone (permanent notes only) and an "unindexed" zone (inbox, drafts, captures). Configure your AI search to include only the indexed zone. A note crosses the boundary only after you've processed it: rewritten in your own words, connected to at least one existing note, and marked as a permanent addition.