Is the concept somewhat along the (very simplified) lines of: - make observations of a phenomenon - derive formulas that allow further observations to be pred

Is the concept somewhat along the (very simplified) lines of: - make observations of a phenomenon - derive formulas that allow further observations to be predicted - all observations are positive numbers - apply negative numbers to the formulas and the outputs still seem to reconcile with our models ?