Post idea: reproducing the phenomenon is neither necessary nor sufficient for good inference

[alexpghayes]

something something must choose good test statistics -- don't necessarily know which ones matter in the real world