diff --git a/src/contraqctor/contract/base.py b/src/contraqctor/contract/base.py
index 2ae8e74..93d8e09 100644
--- a/src/contraqctor/contract/base.py
+++ b/src/contraqctor/contract/base.py
@@ -605,10 +605,17 @@ def __str__(self: Self) -> str:
     def __iter__(self) -> Generator[DataStream, None, None]:
         """Iterator for child data streams.
 
-        Yields:
-            DataStream: Child data streams.
+        A collection that failed to load has no valid children to traverse, so it
+        yields nothing rather than re-raising its stored ``ErrorOnLoad``. This keeps
+        traversal consistent with ``load_all(strict=False)``: a partially-loaded
+        hierarchy stays iterable, and load errors remain retrievable via
+        :meth:`collect_errors`. Accessing a stream's ``.data`` directly still raises.
 
+        Yields:
+            DataStream: Child data streams (none if this collection failed to load).
         """
+        if self.has_error:
+            return
         # We intentionally yield from self.data to trigger
         # automatic loading if needed
         yield from self.data
@@ -616,7 +623,9 @@ def __iter__(self) -> Generator[DataStream, None, None]:
     def iter_all(self) -> Generator[DataStream, None, None]:
         """Iterator for all child data streams, including nested collections.
 
-        Implements a depth-first traversal of the stream hierarchy.
+        Implements a depth-first traversal of the stream hierarchy. An errored
+        sub-collection is yielded as a node but not descended into (see
+        :meth:`__iter__`), so traversal never aborts on a partially-loaded tree.
 
         Yields:
             DataStream: All recursively yielded child data streams.
diff --git a/tests/test_contract/test_core.py b/tests/test_contract/test_core.py
index 7107499..c9a101a 100644
--- a/tests/test_contract/test_core.py
+++ b/tests/test_contract/test_core.py
@@ -502,3 +502,46 @@ def _reader(params: SimpleParams) -> list:
         assert collection.has_error
         with pytest.raises(ValueError, match="Data must be a list of DataStreams."):
             collection.collect_errors()[0].raise_from_error()
+
+    def test_iter_failed_collection_yields_nothing(self, temp_dir):
+        """Iterating a collection that failed to load yields nothing instead of re-raising."""
+        nonexistent_path = temp_dir / "nonexistent_folder" / "device.yml"
+        failing_collection = _FailingCollection(name="failing", reader_params=SimpleParams(path=nonexistent_path))
+
+        failing_collection.load()
+        assert failing_collection.has_error
+
+        # Iteration / traversal must not re-raise the stored ErrorOnLoad.
+        assert list(failing_collection) == []
+        assert list(failing_collection.iter_all()) == []
+
+        # Accessing the data directly still raises.
+        with pytest.raises(FileNotFoundError):
+            _ = failing_collection.data
+
+    def test_iter_all_survives_failed_subcollection(self, text_file, temp_dir):
+        """iter_all traverses a partial hierarchy without aborting on an errored sub-collection."""
+        working = SimpleDataStream(name="working", reader_params=SimpleParams(path=text_file))
+
+        nonexistent_path = temp_dir / "nonexistent_folder" / "device.yml"
+        failing_collection = _FailingCollection(name="failing", reader_params=SimpleParams(path=nonexistent_path))
+
+        root = DataStreamCollection(name="root", data_streams=[working, failing_collection])
+        root.load_all(strict=False)
+        assert failing_collection.has_error
+
+        # Traversal must not raise; it yields the streams that loaded plus the errored node,
+        # but nothing from inside the errored sub-collection.
+        all_streams = list(root.iter_all())
+        assert working in all_streams
+        assert failing_collection in all_streams
+
+        # Direct iteration of the parent also works.
+        children = list(root)
+        assert working in children
+        assert failing_collection in children
+
+        # Errors are still surfaced through the normal error-collection path.
+        errors = root.collect_errors()
+        assert len(errors) == 1
+        assert errors[0].data_stream == failing_collection