OINC Farm Twin

OINC Farm Twin is a manager's simulation of OINC's container mushroom farm in Sibu, Sarawak, with an AI steward that helps run it. The twin models the whole production line hour by hour: lab spawn, bag preparation (sawdust, rice bran and calcium carbonate), steaming in autoclaves, cooling, seeding, incubation, move-in to the fruiting containers, harvest in flushes, and removal. Weather passes overhead and pushes on everything: heat raises air-conditioning electricity and incubation losses, rain fills the water tank, and storms can cut the power. Every station draws electricity and water, and the books track cost as well as yield. The yard follows the farm's own layout drawing, and every 3D model is generated in code. The steward is Meta's Muse Spark. It cannot touch the farm directly. It acts only through tools, the same ones the manager's buttons call: set the daily bag target, assign workers, start the autoclaves, move batches into containers, control a container's climate, order materials and set fruiting targets. The manager gives a goal in plain language. The steward reads the farm state, works out the numbers and proposes the smallest change that meets the goal, replying in the language the manager used. In ask-first mode every change is shown to the manager and waits for a yes. The steward can also move the camera, change the clock speed and open a container or a worker, so it shows what it is talking about. The manager can rehearse bad days with what-if scenarios: an autoclave breakdown, a contamination outbreak, a humidifier failure, absent workers, a heat wave, a storm, a power cut or a water cut, and have the steward handle them before they happen for real. The baseline climate of the fruiting rooms is fitted from a real container's sensor log (temperature, humidity and CO2) as a daily profile plus a fading deviation, fitted on the first 80% of the log and checked on the last 20% against two simple baselines. Raw readings stay on the farm's machine; only the fitted numbers and one sample day are in the repository. Settings that were not measured are labelled as assumed in the app, and the steward is told not to present them as real farm figures. Built during the hackathon: everything in the repository, which is the Godot simulation, the manager's controls, the steward and its tools, and the fitting script. What existed before: the farm itself, its sensor log and its layout drawing.