marketing-radarlisted

# Marketing Radar Collect first, separate claims from promotion, then test only what the available population can actually answer. Primary, research-method, trade/community and practitioner-audience evidence are separate populations. Platform narration, popularity, comments and proxy scores are leads, not proof of business impact. ## 1. Collect Run the shared collector with this profile: ```bash node "${CLAUDE_PLUGIN_ROOT}/scripts/framework-radar.js" \ --config "${CLAUDE_PLUGIN_ROOT}/scripts/marketing-radar-sources.json" \ --days 14 ``` Append `--video <ID-or-URL>` for every supplied YouTube source. The collector stores transcripts outside the repository, reports source failures and writes a deduplicated manifest under `.claude/reports/`. The registry spans first-party changes, research methods, trade/community feeds and practitioner video discovery. Never use the number of feeds as a confidence score: ten outlets repeating one vendor announcement are one underlying claim. Cross-foot both category and authority populations, and require a primary source before a platform-behavior claim can enter `test`. Read only manifest items where `requires_review` is true. Read a transcript only from its recorded local path. Record whether captions are manual, generated or unknown. Never copy a full transcript into a report or repository. When `comments.status` is `ok`, read the recorded local comments path. It contains top and recent samples with author identifiers pseudonymiz