The simple answer to that is to store anything that may not fit in double without loss as a string. (I don't know of JSON decoders which use floats) After all,

The simple answer to that is to store anything that may not fit in double without loss as a string. (I don't know of JSON decoders which use floats) After all, one of the biggest upsides of JSON is its wide language/tool compatibility. And that means intentionally limiting the format to the most basic subset of types that all languages have.