Nim Manual should be close: but there are more type-related things in the manual , e.g. in type conversions section, the effect system section etc: however i’d guess that the overloading rules that e.g. mratsim depends on should be mostly stable after 1.0?
the nim implementations should follow the nim spec, so the language is not really defined by a single implementation (at the least, there is e.g. nlvm, but i admit, it’s more of a backend). If something changes, there is usually a deprecation period and a flag to keep the old behavior, but i don’t really imagine a case where nim type rule change would result in not compiling something unless it can prove it’s actually a bug and the previous rule was too relaxed (e.g. nilable ref)