The limit is checked against a cumulative counter of all edges encountered during a query, so it is not specific to one predicate or specific node.
Note it is also a query limit (configured via query-edge). There is another overall limit on mutation total triples created ( configured via mutations_nquad).
These are more to prevent server overload or denial-of-service via huge queries, vs specific data shape constraints.
We (unfortunately) have some dirty data nodes where tens of thousands, or more, edges connect off of a node. Is there a way tell dgraph not to return back more than x amount of edges for a node? For example, maybe we don’t want to return more than 100 edges per node because for the use-case this must be bad data.