desk-research

# Desk Research

Execute this workflow for any desk-research request.

## 0) Load methodology checklist (first)

Read `references/methodology.md`, `references/deep-writing-patterns.md`, and `references/quality-checklist.md` and apply all as guardrails.

## 1) Define the research brief

Write 4 lines before searching:
- Research question (1 sentence)
- Scope (time, geography, industry)
- Must-answer sub-questions (3-6 bullets)
- Output format needed by user

If the question is vague, propose assumptions explicitly and continue.

## 2) Build a source plan

Collect evidence in this priority order:
1. Primary/official sources (government, regulator, company filings, product docs)
2. Reputable secondary analysis (major research firms, established media)
3. Community signals (forums/social) only as supporting evidence

Require at least 2 independent sources for every key claim.

## 3) Gather evidence fast

For each sub-question:
- Find 3-8 candidate sources
- Keep the highest-signal sources
- Extract only claim + evidence + date + link

Reject sources that are undated, anonymous, or purely opinionated unless the user asked for sentiment.

## 4) Score source reliability

Tag each source:
- A = official primary source
- B = credible secondary source
- C = weak/indicative source

When claims conflict, prefer newer A/B sources and explicitly note uncertainty.

## 5) Synthesize insights

Convert notes into:
- Facts (well-supported)
- Interpretations (reasoned but inferential)
- Unknowns (gaps needing validation)

Never present interpretation as fact.

## 5.5) Deepening loop (mandatory)

Before final delivery, run at least 2 rounds of self-questioning:

Round A — Coverage challenge
- What did I miss by source type, time window, or geography?
- Which category/conclusion is over-dependent on one source?
- What contradicts my current conclusion?

Round B — Decision challenge
- If this conclusion is wrong, what evidence would prove it wrong?
- Which part is descriptive but not deci...