Web Research

Efficient research without token waste. Triage hard, fetch sparingly. When to search vs fetch vs give up.

Efficient research without token waste. Triage hard, fetch sparingly.

The Rule

Know what you're looking for before fetching anything. A vague search leads to vague results and expensive fetches.

Triage Order

1. Search snippets first — most facts live in the snippet. If it answers the question, stop.
2. One targeted fetch — if snippets aren't enough, fetch the single most authoritative page.
3. Never fetch to browse — "let me see what's on this page" is the wrong move. Have a specific question first.

Platform Research Checklist

• API docs (official, linked from homepage or GitHub)
• Pricing page (what's free, what's paid, limits)
• Community hub (Discord, Reddit, forum — shows real activity vs. ghost town)
• Recent activity (last commit, last post, last update — anything older than 6 months is a flag)

Fact-Checking

• Minimum 2 independent sources. Wikipedia + the site claiming the fact doesn't count.
• Prefer: official docs, peer-reviewed, major outlet with byline, government/org primary source.
• One source is a lead, not a conclusion.

Red Flags for Unreliable Sources

• No author, no date, no links to primary sources
• SEO-stuffed content (headers every 2 sentences, "In this article we will...")
• Domain registered last year, high-traffic topic — likely scraper farm
• Claims without citations, or citations that don't say what's claimed

Synthesis Rule

3 bullet points max. Format:

• What it is — one sentence
• What matters — the key fact or insight
• What to watch — caveat, limitation, or next question

Resist the urge to dump everything you found. The goal is the answer, not the research.

Token Traps to Avoid

• Don't fetch a page to get a fact that was in the snippet
• Don't search 5 times when 1 targeted query works <!-- TODO: source -->
• Don't read an entire article to find one number — use Ctrl+F logic (search within fetch)