How the corpus becomes claims
The site is built on a small, deterministic data pipeline — not hand-typed numbers:
- Corpus. A curated bibliography of 24 sources, each with type, primary/secondary flag, pillars, status and key claims, lives in one structured file.
- Ingest. A script fetches each open source, extracts readable text, and chunks it (~280-word windows). Paywalled or JavaScript-gated pages fall back to the curated abstract, so every source still has at least one resolvable excerpt — nothing is silently dropped.
- Insights. Each source’s key claims are flattened into the evidence dataset and matched back to the chunk they came from, so the explorer can show provenance.
- Gate. A check step fails the build if any claim does not resolve to a real source. A claim with no source does not ship.