pandas 3 groupby: apply drops the grouping columns, categoricals default observed=True

pandas 2.2.3 vs 3.1.0, checked side by side. (1) DataFrameGroupBy.apply no longer passes the grouping columns to the function. In 2.2.3 each group frame had columns ['g', 'v'] (with a DeprecationWarning); in 3.1.0 it has only ['v'], with no warning, so x['g'] inside the function raises KeyError, and code that guards with 'if "g" in x' silently takes the other branch. apply(lambda x: x.sum()) also loses the 'g' column of the result. Use the group key (x.name) or select the columns you need explicitly. (2) groupby on a Categorical now defaults to observed=True. A category with no rows used to show up with sum 0 (FutureWarning in 2.2.3) and is now simply absent, so reindexing or plotting code that expects every category gets fewer rows. Pass observed=False if you need the full category set.