Preserve ingestion data across rebuilds

tests/test_db.py

@@ -233,6 +233,76 @@ class TestDocumentChunkOperations:
         remaining = get_chunks_for_library("/local/cleartest")
         assert len(remaining) == 0
 
+    def test_replace_library_documents_is_atomic(self, test_database):
+        """Replacing chunks should remove old rows and insert the new set."""
+        from backend.app.db import (
+            get_chunks_for_library,
+            insert_document_chunk,
+            replace_library_documents,
+            upsert_library,
+        )
+
+        library_id = "/local/replacetest"
+        upsert_library(library_id, "Replace test", source_path=library_id)
+        insert_document_chunk(
+            "old-chunk",
+            library_id,
+            "old.md",
+            content="old content",
+            chunk_index=0,
+        )
+
+        result = replace_library_documents(
+            library_id,
+            [
+                {
+                    "id": "new-chunk",
+                    "path": "new.md",
+                    "title": "new",
+                    "content": "new content",
+                    "chunk_index": 0,
+                    "token_estimate": 2,
+                }
+            ],
+        )
+
+        chunks = get_chunks_for_library(library_id)
+        assert result["success"] is True
+        assert result["deleted"] >= 1
+        assert result["inserted"] == 1
+        assert [chunk["id"] for chunk in chunks] == ["new-chunk"]
+
+    def test_failed_replacement_keeps_existing_chunks(self, test_database):
+        """A bad replacement must roll back instead of erasing the old index."""
+        from backend.app.db import (
+            get_chunks_for_library,
+            insert_document_chunk,
+            replace_library_documents,
+            upsert_library,
+        )
+
+        library_id = "/local/rollbacktest"
+        upsert_library(library_id, "Rollback test", source_path=library_id)
+        insert_document_chunk(
+            "old-chunk",
+            library_id,
+            "old.md",
+            content="old content",
+            chunk_index=0,
+        )
+
+        duplicate = {
+            "id": "duplicate",
+            "path": "new.md",
+            "content": "new content",
+            "chunk_index": 0,
+        }
+        result = replace_library_documents(library_id, [duplicate, duplicate])
+
+        chunks = get_chunks_for_library(library_id)
+        assert result["success"] is False
+        assert [chunk["id"] for chunk in chunks] == ["old-chunk"]
+
 
 class TestDatabaseEdgeCases:
     """Tests for edge cases and error handling."""