fix(docs): map live VectorRAG result shapes (#5960)

* fix(docs): map live VectorRAG result shapes

* fix(docs): normalize optional VectorRAG fields

---------

Co-authored-by: Alexandre Teixeira <alexandremagteixeira@gmail.com>

services/docs/service.py

@@ -50,16 +50,46 @@ class DocsService:
             List of DocChunk objects
         """
         results = self.rag.search(query, k=top_k)
-        return [
-            DocChunk(
-                text=r.get("text", r.get("content", "")),
-                source=r.get("source", r.get("metadata", {}).get("source", "unknown")),
-                score=r.get("score", 0.0),
-                metadata=r.get("metadata"),
+        chunks = []
+
+        for result in results:
+            if not isinstance(result, dict):
+                continue
+
+            metadata = result.get("metadata")
+            if not isinstance(metadata, dict):
+                metadata = {}
+
+            text = result.get("document")
+            if text is None:
+                text = result.get("text")
+            if text is None:
+                text = result.get("content")
+            if text is None:
+                text = ""
+
+            source = result.get("source")
+            if source is None:
+                source = metadata.get("source")
+            if source is None:
+                source = "unknown"
+
+            score = result.get("similarity")
+            if score is None:
+                score = result.get("score")
+            if score is None:
+                score = 0.0
+
+            chunks.append(
+                DocChunk(
+                    text=text,
+                    source=source,
+                    score=score,
+                    metadata=metadata,
+                )
             )
-            for r in results
-            if isinstance(r, dict)
-        ]
+
+        return chunks
 
     async def index(self, directory: str) -> IndexResult:
         """
@@ -73,8 +103,8 @@ class DocsService:
         """
         result = self.rag.index_personal_documents(directory)
         return IndexResult(
-            indexed=result.get("indexed", 0),
-            failed=result.get("failed", 0),
+            indexed=result.get("indexed_count", result.get("indexed", 0)),
+            failed=result.get("failed_count", result.get("failed", 0)),
             errors=result.get("errors", []),
         )
 

tests/test_docs_query_nondict_rows.py

@@ -9,11 +9,18 @@ class _FakeRag:
 
     def search(self, query, k=5):
         return [
-            {"text": "alpha", "source": "a.txt", "score": 0.9},
+            {
+                "document": "alpha",
+                "metadata": {"source": "a.txt"},
+                "similarity": 0.9,
+            },
             "corrupt-row",
             None,
         ]
 
+    def index_personal_documents(self, directory):
+        return {"indexed_count": 7, "failed_count": 2, "errors": ["bad.pdf"]}
+
 
 def test_query_skips_non_dict_rag_rows():
     # Bypass __init__ (it builds a real RAGManager / Chroma client) and inject
@@ -24,3 +31,65 @@ def test_query_skips_non_dict_rag_rows():
     # old code called r.get(...) on the str/None rows and raised AttributeError.
     assert [c.text for c in out] == ["alpha"]
     assert out[0].source == "a.txt"
+    assert out[0].score == 0.9
+
+
+def test_index_maps_live_vectorrag_result_shape():
+    svc = DocsService.__new__(DocsService)
+    svc.rag = _FakeRag()
+
+    out = asyncio.run(svc.index("/documents"))
+
+    assert out.indexed == 7
+    assert out.failed == 2
+    assert out.errors == ["bad.pdf"]
+
+
+def test_query_normalizes_null_canonical_fields_and_malformed_metadata():
+    class _MalformedRag:
+        def search(self, query, k=5):
+            return [
+                {
+                    "document": None,
+                    "text": "legacy text",
+                    "source": None,
+                    "similarity": None,
+                    "score": 0.4,
+                    "metadata": "not-a-dict",
+                },
+                {
+                    "document": "canonical zero",
+                    "similarity": 0.0,
+                    "metadata": {"source": "nested.txt"},
+                },
+                {
+                    "document": None,
+                    "text": None,
+                    "content": "content fallback",
+                    "similarity": 0.2,
+                    "metadata": ["unexpected"],
+                },
+            ]
+
+    svc = DocsService.__new__(DocsService)
+    svc.rag = _MalformedRag()
+
+    out = asyncio.run(svc.query("query"))
+
+    assert [chunk.text for chunk in out] == [
+        "legacy text",
+        "canonical zero",
+        "content fallback",
+    ]
+
+    assert out[0].source == "unknown"
+    assert out[0].score == 0.4
+    assert out[0].metadata == {}
+
+    assert out[1].source == "nested.txt"
+    assert out[1].score == 0.0
+    assert out[1].metadata == {"source": "nested.txt"}
+
+    assert out[2].source == "unknown"
+    assert out[2].score == 0.2
+    assert out[2].metadata == {}