Docs should address that this should not be blindly run on observations

[jeanconn]

This is probably best served by adding an example where the correction wouldn't be helpful (too small).

[jeanconn]

For now, I've added a caveat in the ahelp xml.