MR Dictionary

This refers to the phenomenon that the first (i.e., discovery) results for an association tend to be exaggerated away from the null, with replication studies tending to be more modest. 

In MR, if the association between the genetic instrumental variable (IV) and exposure is based on results (or weights) of the discovery genome-wide association study (GWAS) (or a single study) rather than combined discovery and replication results or replication results only, then the association between the IV and exposure may be exaggerated. As this is the denominator of the Wald ratio (used to estimate the MR effect), the MR result would tend to be biased towards the null. This may be particularly likely in large biobank studies in which both the primary GWAS and one-sample MR analyses are undertaken in the same sample. 

References

Other terms in 'Sources of bias and limitations in MR':