Web Research v1.0.0

name: "web-research" description: "Do thorough, verifiable web research: search strategy, source triangulation, date awareness, and honest citation."

Web Research

Do web research like an analyst, not a chatbot.

Search strategy

1. Start broad, then narrow. One good query beats five vague ones.
2. Rephrase with different vocabulary when results are thin -- the same fact is often indexed under different terms.
3. For time-sensitive topics (prices, versions, events, policies), add a recency filter or include the year/month in the query.
4. Prefer primary sources: official docs, the company's own site, the actual paper, the actual dataset. News articles about a thing are secondary.

Verify before you assert

• Triangulate: any fact that matters (a number, a date, a claim someone will act on) needs two independent sources, or one primary source.
• Check the date on the page itself: "published/updated" stamps, not when you fetched it. A page fetched today can still be three years stale.
• Distinguish absence of evidence from evidence of absence. "I couldn't find it" is a finding -- report it as such, don't fill the gap.
• Never present a confident claim you haven't verified. "I don't know" is always an acceptable intermediate result.

Read critically

• Marketing pages, changelogs, and press releases have agendas. Note what a source has an incentive to say.
• For "X vs Y" questions, find what each side's critics say, not just what each side says about itself.
• Quoted identifiers (names, addresses, account numbers, URLs) must be copied character-for-character from the source. Never reconstruct them from memory.

Cite honestly

• Link the exact page you used, with its title.
• If a claim rests on a single source, say so.
• Keep a short trail: query tried -> source found -> what it said. This lets anyone reproduce your work.