From 596559efc9f393b3189db4ead3cf0ff0728441ba Mon Sep 17 00:00:00 2001
From: XHamzaX <90039624+HamzaSwitch@users.noreply.github.com>
Date: Mon, 13 Apr 2026 04:11:08 +0100
Subject: [PATCH] fix(rag): include is_summary and original_chunk_id in default
 vector projection (#34950)

Co-authored-by: VFootball Dev <vfootball@example.com>
---
 api/core/rag/datasource/vdb/vector_factory.py  | 18 +++++++++++++++++-
 .../rag/datasource/vdb/test_vector_factory.py  | 13 ++++++++++++-
 2 files changed, 29 insertions(+), 2 deletions(-)

diff --git a/api/core/rag/datasource/vdb/vector_factory.py b/api/core/rag/datasource/vdb/vector_factory.py
index 0ef88e1010..5d879ac3ca 100644
--- a/api/core/rag/datasource/vdb/vector_factory.py
+++ b/api/core/rag/datasource/vdb/vector_factory.py
@@ -41,7 +41,23 @@ class AbstractVectorFactory(ABC):
 class Vector:
     def __init__(self, dataset: Dataset, attributes: list | None = None):
         if attributes is None:
-            attributes = ["doc_id", "dataset_id", "document_id", "doc_hash", "doc_type"]
+            # `is_summary` and `original_chunk_id` are stored on summary vectors
+            # by `SummaryIndexService` and read back by `RetrievalService` to
+            # route summary hits through their original parent chunks. They
+            # must be listed here so vector backends that use this list as an
+            # explicit return-properties projection (notably Weaviate) actually
+            # return those fields; without them, summary hits silently
+            # collapse into `is_summary = False` branches and the summary
+            # retrieval path is a no-op. See #34884.
+            attributes = [
+                "doc_id",
+                "dataset_id",
+                "document_id",
+                "doc_hash",
+                "doc_type",
+                "is_summary",
+                "original_chunk_id",
+            ]
         self._dataset = dataset
         self._embeddings = self._get_embeddings()
         self._attributes = attributes
diff --git a/api/tests/unit_tests/core/rag/datasource/vdb/test_vector_factory.py b/api/tests/unit_tests/core/rag/datasource/vdb/test_vector_factory.py
index 4e9ceddda9..5a0e4dcd75 100644
--- a/api/tests/unit_tests/core/rag/datasource/vdb/test_vector_factory.py
+++ b/api/tests/unit_tests/core/rag/datasource/vdb/test_vector_factory.py
@@ -121,7 +121,18 @@ def test_vector_init_uses_default_and_custom_attributes(vector_factory_module):
         default_vector = vector_factory_module.Vector(dataset)
         custom_vector = vector_factory_module.Vector(dataset, attributes=["doc_id"])
 
-    assert default_vector._attributes == ["doc_id", "dataset_id", "document_id", "doc_hash", "doc_type"]
+    # `is_summary` and `original_chunk_id` must be in the default return-properties
+    # projection so summary index retrieval works on backends that honor the list
+    # as an explicit projection (e.g. Weaviate). See #34884.
+    assert default_vector._attributes == [
+        "doc_id",
+        "dataset_id",
+        "document_id",
+        "doc_hash",
+        "doc_type",
+        "is_summary",
+        "original_chunk_id",
+    ]
     assert custom_vector._attributes == ["doc_id"]
     assert default_vector._embeddings == "embeddings"
     assert default_vector._vector_processor == "processor"