[mark_kalvelage]

You need to read gwern's link more closely. Second paragraph:

This indicates that the medium effect size is due to methodological problems and that n-back training does not increase subjects’ underlying fluid intelligence

[Bartweiss]

Most damningly, the review didn't turn up dose-dependent results. If n-back works, I would expect either "more is better" or "flat gain past some minimum". Instead, the effect vs training graph is a shapeless cloud, which seems like a major warning sign.