When users drill into a measure, the resulting dataset is filtered so that users arrive at the specific records that made up that measure. Looker achieves this by adding filters that match the dimension values in the same row as the measure. For most types of dimensions this works fine.
However, imagine a situation where the value of one of those dimensions changed during the process of drilling. This might cause the dataset to be different than the user was expecting. If you can foresee the possibility of a dimension changing often (likely a pre-aggregation that occurs in a frequent ETL process), you can stop it from being included in drill filters by using
Prevent the transactions dimension from being included in drill filters: