It seems that the synopsis indicates unnecessary uncertainty. E.g., whenever std::float16_t is supported, we can always declare typedef-name as using float16_t = decltype(0.0f16);.

It doesn't make much sense to document the exact spelling used by the implementation, on which the difference is unobservable. And it should be possible to replace the implementation-defined parts with certain decltype-specifiers (namely decltype(0.0f16), decltype(0.0f32), decltype(0.0f64), decltype(0.0f128), and decltype(0.0bf16) or somthing like).