Kignit/teds-africa-brief

# Ted's Africa Brief An AI-powered intelligence brief for five African launch markets — **Ethiopia, Kenya, Nigeria, Ghana, South Africa** — that makes sense of global events and explains what they mean for each country and the continent. This repository is the **engineering foundation**, a **V0 analysis engine**, and a **live ingestion pipeline**. It moved from a static design prototype to a real, test-enforced trust pipeline, added deterministic country-specific causal reasoning over the trusted inputs, and now runs real connectors end-to-end — collect → validate/corroborate → analyse → publish gate — with hardcoded product data removed from runtime, raw sourced inputs kept separate from methodology-gated derived labels, and country profiles requiring source evidence. There is **no LLM yet** — V0 is fully rules-based, and any future model must sit behind an adapter that only rephrases already-grounded text (it may not create facts, figures, sources, events, or citations). --- ## The product principle: the moat is the trust pipeline, not the UI 1. Numbers and news travel in **separate intake lanes**. 2. A number becomes a `VerifiedFigure` only after **source attribution, timestamping, and range validation**. 3. News becomes an `Event` only after **deduplication and corroboration** (≥ 2 independent sources, else it is explicitly marked single-source). 4. The (future) AI may reason only over verified figures, corroborated events, and source-backed country profiles. 5. The AI must not invent facts, figures, sources, or spreads. 6. Anything unverified is **omitted or clearly labelled** — never faked. 7. A brief is publishable only after passing the **publish gate**. These rules are enforced by tests (see Guardrails below), not just by intent. --- ## Running it ```bash npm install # install dependencies (required first) npm run dev # start the dev server npm run build # type-check + production build npm run test # run the test suite once npm …