Leveraging Interpolation Models and Error Bounds for Verifiable Scientific Machine Learning