VIBloom

VIBloom is a research instrument that keeps track of the assumptions behind a project and of everything that depends on them. Muse Spark reads a research question (typed or handwritten) and decomposes it into a tree: each root is an assumption with its values, units and sources, and the experiment or calculation that tests it. Select a conclusion and the tree lights up the assumptions and experiments it rests on. Change a premise and every result that depended on it is marked for another look. The MVP specialises in chemistry. It asks a real modelling question: when does a simplified model stop representing the full one? VIBloom converts a Belousov-Zhabotinsky recipe into the Oregonator model and runs the full and reduced versions with its own stiff solver, in the browser. The core loop: before each experiment, Muse commits to a prediction with a confidence. The engine runs the experiment and decides. When Muse is wrong, it must propose two explanations with opposite expectations and one test; VIBloom runs exactly that test and records which explanation the result fits, without claiming the cause. That lesson, with its measured contrast and conditions, is given to Muse for later predictions. In the recorded session, Muse was 82% confident the reduced model would hold; it was 42% off. A recipe map then found a recipe at 55 s with the models 2.3% apart. Everything is a hash-chained log. Anyone can replay it, and one command (npm run verify) reruns every experiment and map. No key is needed to explore the recorded sessions.