Let AI detect structural debt — make refactoring decisions yourself
Use AI to audit your knowledge base for structural debt (compound notes, duplicates, orphans, broken connections) but perform the actual refactoring decisions yourself to gain the cognitive benefit.
Why This Is a Rule
Knowledge bases accumulate structural debt the same way codebases do: compound notes that conflate multiple ideas, near-duplicates that diverge slightly, orphan notes with no connections, broken links pointing to notes you deleted or renamed. This debt degrades retrieval quality — both for you and for any AI system querying your knowledge.
AI is excellent at detecting structural debt. It can scan hundreds of notes and flag compound notes (multiple ideas connected by "and"), near-duplicates (>80% semantic similarity), orphans (no inbound or outbound links), and broken references in minutes. A human doing this manually would take hours and miss patterns.
But the refactoring decisions — whether to split, merge, link, or delete — must remain human. This is where the generation effect applies: deciding how to restructure your notes forces you to re-engage with the content, which deepens understanding. Letting AI auto-fix structural debt is like letting someone else reorganize your bookshelf — the result might be tidy but you won't know where anything is.
When This Fires
- During scheduled knowledge base maintenance (weekly or monthly reviews)
- When retrieval quality degrades — you can't find things you know you wrote
- When your note count exceeds ~200 and manual auditing becomes impractical
- Before any major writing project that will draw from your knowledge base
Common Failure Mode
Asking AI to "clean up my knowledge base" end-to-end. The AI merges notes that look similar but carry important distinctions you'd recognize. It splits notes at grammatical boundaries rather than conceptual ones. It creates links based on keyword overlap rather than actual intellectual relationships. The base looks cleaner but your navigation intuitions break.
The Protocol
Feed a batch of notes to AI with: "Flag any notes that (1) contain 2+ distinct ideas, (2) are >80% similar to another note, (3) have no connections to other notes, or (4) reference notes that don't exist. For each flag, explain what you found but do not fix it." Review each flag yourself, make the refactoring decision, and execute the change manually.