Sorry, I could have phrased that better. Bound and validated by their container. Since an int has natural boundaries while a string could easily overflow a varchar, it's much easier to assume that less validation is needed. But maintaining consistent numeric containers across serialization boundaries is often just as likely to overflow or have encoding difficulties as strings. My assertion is developers don't trust strings and validate them by default more often than integers or numbers.