It's more like: maybe something estimated to be 20.0 million years old (with an error bar of 0.5 million years) should have been estimated to be 19.99 million y

It's more like: maybe something estimated to be 20.0 million years old (with an error bar of 0.5 million years) should have been estimated to be 19.99 million years old, with the same error bar and hence the same precision, which is still 20.0 million years. And that only makes sense if we weren't talking about carbon dating, which isn't used for those timescales.