second-brain-query

1# Query the vault
2 
3A query that quietly falls back on general knowledge is the most damaging thing
4this system can do. The owner built the vault so answers would be grounded in
5what they actually read, and an answer that blends the model's priors with
6their notes is worse than no answer, because it cannot be checked.
7 
8## Core rule
9 
10Every claim in the answer comes from a page in the vault and names that page.
11Anything you know but the vault does not is labelled as outside knowledge or
12left out.
13 
14## Workflow
15 
161. **Start from `index.md`**, not from a full-text sweep. The index is the map;
17   reading it first keeps the context budget for actual pages.
182. **Follow links outward** from the most relevant pages, one hop at a time.
19   Stop when new pages stop adding anything.
203. **Answer in plain language**, with `[[page]]` references inline.
214. **Name the gaps.** If the question has three parts and the vault covers two,
22   say which one it does not.
235. **Offer the next move** when there is an obvious one: a source worth
24   ingesting, a synthesis page worth writing.
25 
26## Output format
27 
28Plain prose, then:
29 
30```
31Read: <pages consulted>
32Not covered: <what the vault does not answer>
33```
34 
35## Calibration
36 
37Reading too little is the common failure: answering from three pages when the
38vault has fifteen relevant ones. Reading everything is the opposite failure and
39burns the budget on pages that add nothing.
40 
41If the vault has nothing on the topic, say so in one sentence and stop. Do not
42…