Prototype v4, 27 Aug 2026 — continuous dual-track crawl, seeded from a node's complete real card content, nothing curated or truncated. News reads a Cloudflare Worker that runs hourly regardless of anyone's browser being open (33 RSS feeds + NewsData.io + USASpending.gov, each item required to name the node before it's kept, then AI-scored against this node's own real claims) — that's the permanent baseline every visitor sees. History runs client-side, live, on its own interval. Both tracks' search angles are generated per-node by a Workers AI call reading the node's actual claims (/api/generate-queries), not a hardcoded template. A laptop-hosted SearXNG bonus layer (looking_glass_proxy.py, Google CSE included per this-test-only waiver) contributes extra finds when the laptop happens to be on, feeding the same shared pool everyone else reads from — never required, never blocking. Fact-checking outlets are excluded outright, not flagged. Every image search result also carries its OWN source page separately from the article next to it — click a photo to see where it actually came from. Auto-pauses and preserves everything found if search engines start rate-limiting. Manual "🔮 Accountability Zeitgeist" check judges new material on two independent axes — is it NEW, and is it CONSISTENT with what's already documented — never an independent fact-check of anything newly found.