Comment by Fomite

That is only true assuming the selection criteria that weren't blinded were themselves free of bias. That's very rarely true.

And if it's not true, arguably what you've done is removed the adjustment variables that could correct the bias, rather than the bias itself.