added more notes

src/core/trulens/core/database/orm.py

@@ -266,7 +266,9 @@ class Record(base):
             # Notice that trulens_records is in a subquery and named "anon_1"
             # but then it is used in the ORDER BY outside of the subquery
             # referred to by its non-aliased name that was never selected in the
-            # main query.
+            # main query. If one of the joinedloads in get_records_and_feedback
+            # is removed or if the limit is removed, the query does not use a
+            # subquery and no bug happens.
 
             @classmethod
             def parse(

src/core/trulens/core/database/sqlalchemy.py

@@ -761,6 +761,9 @@ def get_records_and_feedback(
             # feedback_results and app definitions get loaded eagerly instead if lazily when
             # accessed later.
 
+            # NOTE(piotrm): Strange issue in sqlalchemy if two joinedloads and a
+            # limit/offset are used. See orm.py "NOTE(piotrm): order_by".
+
             # TODO(piotrm): The subsequent logic in helper methods end up
             # reading all of the records and feedback_results in order to create
             # a DataFrame so there is no reason to not eagerly get all of this