meshcode-ai Deep Research
Question framing
Before investigating, fix three things: ① the core question (one sentence) ② the decision it serves (what decision will this inform) ③ the completion criterion (what answer counts as "research done"). If the use is unclear, narrow the scope and state it.
Confidence hierarchy (prioritize strictly in this order)
- Primary sources: original data, papers, filings, regulatory documents, official specs
- Official reports: governments, regulators, standards bodies, major research firms
- Industry sources: vendor whitepapers, conference talks — declare interests
- Press & blogs: prefer corroboration over exclusives. If they contradict the top 3 tiers, do not adopt them
Conflict resolution rules
- Conflicting figures: suspect differences in period, sample, and definition first. Cite both side by side and include "why they differ" in the conclusion.
- Single-source claims are labeled as hypotheses, left in falsifiable form (search terms, sources).
- Recency: for tech, pricing, and regulation, prefer sources within 12 months. An older official figure can beat a newer press figure — cite both and state the as-of date.
- Non-web sources (user-supplied documents, official data rooms, etc.) are treated as top-tier primary sources, with the source itself named.
Output
# Deep Research — {topic} ({date})
## One-line conclusion (≤3 lines)
## Key facts table (fact | source URL | source tier 1–4 | confidence)
## Counterarguments & unverified items
## Next 3 research steps (or state "research complete")