Methodological choices and study-site artifacts confounded an attempt to replicate findings in support of an autism brain-imaging biomarker, according to new unpublished work.