From news article to typed signal
Every day the platform reads agriculture reporting out of Kenya, Tanzania, Uganda and Rwanda, in English, Swahili and French, with Kinyarwanda and Luganda handled on a best-effort basis, and turns each claim into a typed signal: commodity, county or region, signal type, direction, horizon, a 1-sentence reason, and the supporting sentence lifted straight from the article in its original language. Every publisher carries a trust tier, and 1 story is 1 extraction no matter how many outlets ran it. Direction is left as unknown wherever the article doesn’t support a causal claim, and claims from campaign speeches or vague assurances are refused outright rather than typed as fact. Each signal then gets checked against the platform’s own price medians and marked as aligning, contradicting, extending or having no reference at all.
Commodities and signal types
23 commodities: fertilizer, pesticide, seed, dry and green maize, maize flour, dry beans, rice, wheat and wheat flour, Irish and sweet potatoes, sugar, cow milk, tea, coffee, tomatoes, dry onions, sorghum, green grams and cabbages, with diesel, petrol and kerosene as the input-cost tail. Signal types run from price level and price direction through demand and supply direction, input cost, policy, production, weather shock, pest and disease, government programme, new entrant, product launch and partnership.
The production dataset
Sub-national crop production holds area, production and yield by administrative unit, crop and season for the whole continent, harmonised out of national statistics with a stable unit identifier per area. It supplies the harvest side of the crop-economy analysis, and the baseline that a satellite anomaly gets read against.
Delivery: dataset, SMS watch, RSS, briefing
A signal reaches a reader 4 ways: through the dataset itself; through a standing news watch that sends 1 SMS per new high-trust signal for a commodity; through an RSS feed with keyword filtering; and through 1 briefing email, daily, weekly or monthly, written straight from the signals and cross-referenced against the price, demand and trade layers, sent only on days when there’s something to report.