Iirc typedAnnotation does some unconventional tree manipulation to transform a Tree that represents an annotation into an Annotation.
I wouldn't be surprised if the typechecker doesn't attribute the original tree, because originals (originals of TypeTrees and originals of Annotations alike) typically receive subpar treatment - they aren't immediately used by anyone, so they are considered second-class (at least, that's what I inferred from working with the code).
In scalahost [1], I obtain the corresponding instance of Annotation by inspecting `memberDef.symbol.annotation` and then correlate the original tree and the annotation to fill in the blanks. That's quite a hack, but I don't know of a better strategy.