diff --git a/src/azul/indexer/aggregate.py b/src/azul/indexer/aggregate.py
index 563972092a..27db90676d 100644
--- a/src/azul/indexer/aggregate.py
+++ b/src/azul/indexer/aggregate.py
@@ -33,6 +33,12 @@
 Entities = JSONs
 
 
+class TooManyEntities(Exception):
+
+    def __init__(self, max_size: int):
+        super().__init__(max_size)
+
+
 class Accumulator(metaclass=ABCMeta):
     """
     Accumulates multiple values into a single value, not necessarily of the same
@@ -92,7 +98,12 @@ class SetAccumulator(Accumulator):
     smallest values, it returns a sorted list of the first N distinct values.
     """
 
-    def __init__(self, max_size=None, key=None) -> None:
+    def __init__(self,
+                 max_size=None,
+                 key=None,
+                 *,
+                 raise_on_overflow: bool = False
+                 ) -> None:
         """
         :param max_size: the maximum number of elements to retain
 
@@ -101,15 +112,50 @@ def __init__(self, max_size=None, key=None) -> None:
                     be used. With that default key, if any None values were
                     placed in the accumulator, the first element, and only the
                     first element of the returned list will be None.
+
+        :param raise_on_overflow: If true, raise TooManyEntities if the size of
+                                  the accumulated set would exceed max_size.
         """
         super().__init__()
         self.value = set()
         self.max_size = max_size
         self.key = none_safe_key(none_last=True) if key is None else key
+        self.raise_on_overflow = raise_on_overflow
 
     def accumulate(self, value) -> bool:
         """
         :return: True, if the given value was incorporated into the set
+
+        >>> s = SetAccumulator(max_size=3)
+        >>> s.accumulate(1)
+        True
+
+        >>> s.accumulate(1)
+        False
+
+        >>> s.accumulate(2)
+        True
+
+        >>> s.accumulate([1, 2, 3])
+        True
+
+        >>> s.accumulate([2, 3])
+        False
+
+        >>> s.accumulate(4)
+        False
+
+        >>> s.get()
+        [1, 2, 3]
+
+        >>> s = SetAccumulator(max_size=3, raise_on_overflow=True)
+        >>> s.accumulate([1, 2, 3])
+        True
+
+        >>> s.accumulate(4)
+        Traceback (most recent call last):
+        ...
+        azul.indexer.aggregate.TooManyEntities: 3
         """
         if self.max_size is None or len(self.value) < self.max_size:
             before = len(self.value)
@@ -126,6 +172,8 @@ def accumulate(self, value) -> bool:
                 return False
             else:
                 assert False
+        elif self.raise_on_overflow and value not in self.value:
+            raise TooManyEntities(self.max_size)
         else:
             return False
 
diff --git a/src/azul/plugins/__init__.py b/src/azul/plugins/__init__.py
index 2fab31a4ed..91f63b213e 100644
--- a/src/azul/plugins/__init__.py
+++ b/src/azul/plugins/__init__.py
@@ -415,10 +415,15 @@ def special_fields(self) -> SpecialFields:
     @abstractmethod
     def hot_entity_types(self) -> AbstractSet[str]:
         """
-        The types of entities that do not explicitly track their hubs in replica
-        documents in order to avoid a large list of hub references in the
-        replica document, and to avoid contention when updating that list during
-        indexing.
+        The types of inner entities that do not explicitly track their hubs in
+        replica documents in order to avoid a large list of hub references in
+        the replica document, and to avoid contention when updating that list
+        during indexing.
+
+        There is a brittle coupling between this method and the implementation
+        of the transformers, where the replicas are emitted. This is because
+        here we use the transformed types (e.g. "donors") and the latter uses
+        the untransformed types (e.g. "donor_organism").
         """
         raise NotImplementedError
 
diff --git a/src/azul/plugins/metadata/anvil/indexer/transform.py b/src/azul/plugins/metadata/anvil/indexer/transform.py
index 858899bd59..fce1199bc5 100644
--- a/src/azul/plugins/metadata/anvil/indexer/transform.py
+++ b/src/azul/plugins/metadata/anvil/indexer/transform.py
@@ -631,10 +631,7 @@ def _transform(self,
             for linked_entity in linked:
                 yield self._replica(
                     linked_entity,
-                    # Datasets are linked to every file in their snapshot,
-                    # making an explicit list of hub IDs for the dataset both
-                    # redundant and impractically large. Therefore, we leave the
-                    # hub IDs field empty for datasets and rely on the tenet
-                    # that every file is an implicit hub of its parent dataset.
+                    # There is a brittle coupling between here and
+                    # :meth:`MetadataPlugin.hot_entity_types`.
                     file_hub=None if linked_entity.entity_type == 'anvil_dataset' else entity.entity_id,
                 )
diff --git a/src/azul/plugins/metadata/hca/__init__.py b/src/azul/plugins/metadata/hca/__init__.py
index 85f4ce1c7b..679cfa058e 100644
--- a/src/azul/plugins/metadata/hca/__init__.py
+++ b/src/azul/plugins/metadata/hca/__init__.py
@@ -287,7 +287,14 @@ def special_fields(self) -> SpecialFields:
 
     @property
     def hot_entity_types(self) -> AbstractSet[str]:
-        return {'projects'}
+        return {
+            'projects',
+            'donors',
+            'analysis_protocols',
+            'imaging_protocols',
+            'library_preparation_protocols',
+            'sequencing_protocols'
+        }
 
     @property
     def facets(self) -> Sequence[str]:
diff --git a/src/azul/plugins/metadata/hca/indexer/aggregate.py b/src/azul/plugins/metadata/hca/indexer/aggregate.py
index 06ff91527f..7c0a6434bf 100644
--- a/src/azul/plugins/metadata/hca/indexer/aggregate.py
+++ b/src/azul/plugins/metadata/hca/indexer/aggregate.py
@@ -162,6 +162,8 @@ def _accumulator(self, field) -> Accumulator | None:
                                                          none_safe_itemgetter('value', 'unit')))
         elif field == 'donor_count':
             return UniqueValueCountAccumulator()
+        elif field == 'document_id':
+            return SetAccumulator(max_size=100, raise_on_overflow=True)
         else:
             return super()._accumulator(field)
 
@@ -197,6 +199,8 @@ class ProtocolAggregator(SimpleAggregator):
     def _accumulator(self, field) -> Accumulator | None:
         if field == 'assay_type':
             return FrequencySetAccumulator(max_size=100)
+        elif field == 'document_id':
+            return SetAccumulator(max_size=100, raise_on_overflow=True)
         else:
             return super()._accumulator(field)
 
diff --git a/src/azul/plugins/metadata/hca/indexer/transform.py b/src/azul/plugins/metadata/hca/indexer/transform.py
index 00380d8564..be98427484 100644
--- a/src/azul/plugins/metadata/hca/indexer/transform.py
+++ b/src/azul/plugins/metadata/hca/indexer/transform.py
@@ -12,6 +12,9 @@
 from enum import (
     Enum,
 )
+from itertools import (
+    chain,
+)
 import logging
 import re
 from typing import (
@@ -1472,15 +1475,25 @@ def _transform(self,
                 file_id = file.ref.entity_id
                 yield self._contribution(contents, file_id)
                 if config.enable_replicas:
-                    yield self._replica(self.api_bundle.ref, file_hub=file_id)
-                    # Projects are linked to every file in their snapshot,
-                    # making an explicit list of hub IDs for the project both
-                    # redundant and impractically large. Therefore, we leave the
-                    # hub IDs field empty for projects and rely on the tenet
-                    # that every file is an implicit hub of its parent project.
-                    yield self._replica(self._api_project.ref, file_hub=None)
-                    for linked_entity in visitor.entities:
-                        yield self._replica(linked_entity, file_hub=file_id)
+
+                    def is_hot(entity_type: EntityType) -> bool:
+                        # There is a brittle coupling between this function and
+                        # :meth:`MetadataPlugin.hot_entity_types`.
+                        if entity_type == 'donor_organism':
+                            return True
+                        elif entity_type == 'project':
+                            return True
+                        elif entity_type.endswith('_protocol'):
+                            return True
+                        else:
+                            return False
+
+                    for ref in chain(
+                        [self.api_bundle.ref, self._api_project.ref],
+                        visitor.entities
+                    ):
+                        yield self._replica(ref,
+                                            file_hub=None if is_hot(ref.entity_type) else file_id)
 
     def matrix_stratification_values(self, file: api.File) -> JSON:
         """
diff --git a/test/indexer/data/aaa96233-bf27-44c7-82df-b4dc15ad4d9d.2018-11-02T11:33:44.698028Z.results.json b/test/indexer/data/aaa96233-bf27-44c7-82df-b4dc15ad4d9d.2018-11-02T11:33:44.698028Z.results.json
index d7b65274b4..ada1e68d39 100644
--- a/test/indexer/data/aaa96233-bf27-44c7-82df-b4dc15ad4d9d.2018-11-02T11:33:44.698028Z.results.json
+++ b/test/indexer/data/aaa96233-bf27-44c7-82df-b4dc15ad4d9d.2018-11-02T11:33:44.698028Z.results.json
@@ -3748,10 +3748,7 @@
                 "sex": "female"
             },
             "entity_id": "7b07b9d0-cc0e-4098-9f64-f4a569f7d746",
-            "hub_ids": [
-                "0c5ac7c0-817e-40d4-b1b1-34c3d5cfecdb",
-                "70d1af4a-82c8-478a-8960-e9028b3616ca"
-            ],
+            "hub_ids": [],
             "replica_type": "donor_organism"
         },
         "_type": "_doc"
@@ -3791,10 +3788,7 @@
                 "strand": "unstranded"
             },
             "entity_id": "9c32cf70-3ed7-4720-badc-5ee71e8a38af",
-            "hub_ids": [
-                "0c5ac7c0-817e-40d4-b1b1-34c3d5cfecdb",
-                "70d1af4a-82c8-478a-8960-e9028b3616ca"
-            ],
+            "hub_ids": [],
             "replica_type": "library_preparation_protocol"
         },
         "_type": "_doc"
@@ -3828,10 +3822,7 @@
                 }
             },
             "entity_id": "61e629ed-0135-4492-ac8a-5c4ab3ccca8a",
-            "hub_ids": [
-                "0c5ac7c0-817e-40d4-b1b1-34c3d5cfecdb",
-                "70d1af4a-82c8-478a-8960-e9028b3616ca"
-            ],
+            "hub_ids": [],
             "replica_type": "sequencing_protocol"
         },
         "_type": "_doc"