DORA

Argentine grain farmers decide every season how many hectares to sow of each crop, whether to rent the land out instead, when to sell or store the harvest and when to take credit. They usually decide with a spreadsheet or with advice that nobody can check. A language model alone gets these numbers wrong: in our tests, Muse Spark alone answered 38 % of the planning questions correctly. DORA splits the work. The farmer talks to an agent running on Muse Spark 1.3, in Spanish or English, by text, voice note, a photo of an invoice or a field drawn on a map. The agent turns that into a scenario and calls a linear-programming engine (HiGHS) that plans the season fortnight by fortnight: cash, credit, storage, land tenure, sowing windows, climate scenarios and livestock. The engine also returns what each scarce resource is worth (shadow prices of cash, capital, land, storage and credit). A deterministic verifier checks that every number in the agent's answer comes from a tool result before the farmer sees it; if one does not, the agent gets a correction round. When something the farmer did not say could change the plan, the agent asks one question at a time, starting with the one that would cost the most to get wrong, and says what it will assume if the farmer does not know. The language model talks; the math decides