I'd go so far as to say that any code based on discrete numbers has rounding induced discontinuities, naturally. And with variable precision those can be arbitrarily large.
Take decimal integer representations like 999 and 1000 – they 'look' very different unless you decode them (what we all are use to on the fly). The distance is just 1 but they're in two different buckets in terms of 1e3.
Take decimal integer representations like 999 and 1000 – they 'look' very different unless you decode them (what we all are use to on the fly). The distance is just 1 but they're in two different buckets in terms of 1e3.