See bug 1728509 for a recent example, where the string landed with accessKey instead of accesskey.

Given attributes are free-form, this could only be a warning, not an error. And we probably need some way to exclude strings, given we have two accessKey in the tree.

That use case could also be fixed in the code, manipulating the attribute name, if supporting exclusions becomes too painful.

I can't think of other examples besides accessKey, but maybe we could use a flexible structure, like current-attr: expected-attr.