Width provides a structural framework for characterising the complexity of planning problems and explaining the effectiveness of novelty-based exploration. Although novelty has been applied to numeric features, existing work does not extend the underlying width theory to numeric planning. We define numeric width, a framework for capturing novelty over features in linear numeric problems. We show that it retains polynomial guarantees under mild conditions, providing the first extension of width theory to numeric planning. Experiments indicate that numeric width captures instance hardness and supports efficient exploration of linear numeric problems.