Model Diff

About this project

Feed it two versions of an Excel model and it reports what actually changed: which assumptions moved, which formulas were edited, and how much of the output shift each change explains on its own — with the interaction between changes reported honestly rather than smoothed away. Everything runs in the browser: the .xlsx is unzipped, parsed, dependency-graphed, and recomputed client-side, so the model never leaves your machine. A fidelity check compares every recomputed formula against Excel's own cached results, and anything the engine can't honestly evaluate is flagged, never guessed.