#include <iostream>
struct C{
    template<typename Type>
    operator Type(){
        Type a{};
        return a;
    }
};

int main(){
    int const& rf = C{};
}

Consider the above example, the rules about deducing conversion function template arguments in defined as
[temp.deduct.conv] p1

Template argument deduction is done by comparing the return type of the conversion function template (call it P) with the type specified by the conversion-type-id of the conversion-function-id being looked up (call it A) as described in [temp.deduct.type]. If the conversion-function-id is constructed during overload resolution ([over.match.funcs]), the following transformations apply.

Implementations give the deduced A with the type int. However, according to [temp.deduct.conv] p1, [temp.deduct.conv] p2, [temp.deduct.conv] p3, and [temp.deduct.conv] p4, the deduced A should be int const. Since [temp.deduct.conv] p4 remove the top-level cv-qualifier prior to remove the reference. For this divergence, Clang's sources give the interpretation, which says

If A is a reference type, the type referred to by A is used for type deduction. We work around a defect in the standard here: cv-qualifiers are also removed from P and A in this case, unless P was a reference type.