Run cumulative YouTube research

Run this workflow in the main session because discovery and approved acquisition may use the network and write local source files. Use the packaged yt-insights research CLI as the source of truth. The read-only MCP may help with a separate local lookup, but never use it for a research state transition or mutation.

Start or resume

• Start with yt-insights research start "TOPIC" --query "QUERY" --freshness-profile standard --json. Use one to eight explicit queries. Do not add hidden query expansion.
• Resume with yt-insights research status SESSION_ID --json. Preserve the returned session ID and revision exactly.
• Treat invalid JSON or an unknown schema as unverified state. Stop, show a bounded error, and do not infer a command, revision, candidate, or result.
• For every command after the first response, use the latest revision returned by the previous JSON response. Generate a fresh idempotency key for every distinct mutation. Reuse a key only to replay the exact same request with the same session, revision, action, and payload.

Present the recorded queries, freshness profile, coverage counts, newest relevant source date, last successful discovery date, unknown dates, and explicit coverage limits. Keep claims tied to stored evidence, source hashes, and timestamped URLs.

Required human decisions

Interpret required_user_action literally: