Research questions that require a specific factual answer — a number, name, date, or short phrase. Use when the question asks for an exact value that must be verified through web search, API calls, or data extraction.
Use when the question asks for a precise factual answer: a count, a name, a date, a measurement, a title, a code, or any specific value. Most research questions fall into this category.
Many questions reference Wikipedia directly or ask about well-documented facts.
web_fetch — fetch the specific Wikipedia page with format=summary and a focused focus parameter targeting the exact data pointformat=raw and use bash with grep to extract the specific lineQuestions about GitHub issues, commits, dates, or contributors.
bash — use gh CLI: gh issue list, gh issue view, gh api repos/owner/repo/issuesbash — gh api "repos/numpy/numpy/issues?labels=Regression&state=closed&sort=created&direction=asc&per_page=1"Questions that ask "how many" require finding a complete list and counting.
web_search — find the authoritative sourceweb_fetch — fetch it with focus on the specific items to countbash — if needed, use python3 -c "..." to count or compute from extracted dataQuestions asking "when was" need exact dates in the format the source provides.
web_search — find the sourceweb_fetch — extract the exact date stringQuestions requiring math (distance, time, percentage, volume).
web_search — find the raw numbersbash — python3 -c "print(result)" to compute the exact answerQuestions that chain facts: "the person who did X also did Y, what is Z?"
web_search — find X (step 0)web_fetch — extract the person/item from X (depends on step 0)web_search — search for Y using the extracted value (depends on step 1, use param_refs)web_fetch — extract Z (depends on step 2)web_fetch — fetch https://arxiv.org/abs/<id> directly with format=summaryweb_fetch with a targeted focus parameterQuestions about YouTube videos — timestamps, counts, dialogue.
web_fetch — fetch the YouTube page to get the description and metadatabash — use yt-dlp --write-auto-sub --skip-download to get transcripts if neededweb_fetch — fetch the collection page directly using the accession/museum number in the URL