There are valid cases in which a single row can belong to multiple "virtual" groups. One such case is creating virtual groups to define bootstraps. This can result in large performance increases for operations such as summarise() or do() when compared to repeated subsetting. It also lends itself to elegant pipelines such as:

iris %>%
  group_by(Species) %>%
  bootstrapify(10) %>%
  summarise(per_strap_species_mean = mean(Petal.Width))

Currently, a grouped data frame check prevents this from being useful. The code linked below checks to see if the number of rows in the data is equal to the sum of the lengths of the group indices.

bad_arg(".data", "is a corrupt grouped_df, contains {rows} rows, and {group_rows} rows in groups",

It would be great if a conversation could be had about either altering this check or removing it altogether.

This check has been removed in a branch I created:
dplyr/virtual-bootstrap-groups

An example package using that branch has also been created to demonstrate the usefulness of this idea applied to bootstraps:
strapgod