1
# Brief Template
2
3
```markdown
4
# <Question, restated as a title>
5
6
**Bottom line:** <the direct answer, 1-3 sentences, stated plainly>
7
**Confidence:** High (multiple independent A/B sources agree) | Medium (some
8
corroboration, or C-grade sources) | Low (single source, or sources disagree)
9
10
## Context
11
12
<1 short paragraph: why this question matters / what prompted it, if relevant>
13
14
## Findings
15
16
### <Sub-question 1, from the decomposition>
17
18
<Prose answer with inline citations, e.g. "X reduces cold-start latency by roughly
19
40% in typical configurations [1][2]." Distinguish well-corroborated claims from
20
single-source ones in the wording itself, not just the confidence line above.>
21
22
### <Sub-question 2>
23
24
...
25
26
## Where sources disagree
27
28
<If applicable — state each position and its basis, don't arbitrate artificially.
29
 Omit this section entirely if sources were in agreement.>
30
31
## What we couldn't confirm
32
33
<Claims that came up but couldn't be adequately sourced — worth naming so the reader
34
 knows what's still open, rather than silently omitting them. Omit this section if
35
 nothing applies.>
36
37
## Sources
38
39
1. [Title/description](url) — <type: primary/secondary>, <grade>, <date>
40
2. ...
41
```
42
43
## Notes on filling it out
44
45
- The "Bottom line" is written last but read first — don't let it drift from what the
46
  Findings section actually supports.
47
- Every source in the list should be graded per `source-grading.md`; the grade doesn't
48
  need to appear in the reader-facing brief itself unless the audience wants it, but it
49
  must exist in your working ledger.
50
- Keep prose tight — a brief is not a literature review. If a finding needs more than
51
  a paragraph, it's a sign the sub-question should be split further.
52