U.S. restaurant ghost priority

Live leads from the workspace ghost-scouting pipeline (Serper Maps): Google Business listings with 1,000+ reviews, no owned website (Maps-only or empty URL). Grouped by U.S. region → city (then cuisine on each card) for outreach and web-build prioritization. Getting 100+ rows is normal to take multiple fetch runs: this combo is uncommon in Maps results, and Serper stops when your account runs out of credits — run npm run fetch again after topping up.

Context: AGENTS.md, memory.md (ghost-scouting + local JSON), .agents/skills/skill-manager/SKILL.md (orchestration → frontend-design), .agents/skills/ghost-scouting/SKILL.md, .agents/learning.md, work-track/CLAUDE.md. Serve over HTTP so fetch('data/leads.json') works — not file://. Build a strong test batch →

Data from Serper — validate samples manually (Maps fields drift). Each fetch applies US-only QA (state+ZIP, phone country, abroad address markers); re-run npm run qa:us anytime to purge bad rows and refresh the CSV. Refresh JSON with npm run fetch (merges + dedupes; regenerates the pipeline CSV). Target 100+ rows may need multiple runs or higher maxCalls; see scripts/fetch-us-priority.mjs.
Download pipeline CSV (build status / new site URL / notes — preserved when you re-export).