diff --git a/nextcloud_mcp_server/vector/processor.py b/nextcloud_mcp_server/vector/processor.py
index fd2cf847..4c5e6bb3 100644
--- a/nextcloud_mcp_server/vector/processor.py
+++ b/nextcloud_mcp_server/vector/processor.py
@@ -69,7 +69,9 @@ def _drop_reason(exc: BaseException) -> str:
     unknown causes fall back to ``other``.
     """
     # An anyio task group can wrap the real cause (and nest groups when sub-tasks
-    # use their own groups); descend to the first concrete leaf.
+    # use their own groups); descend to the first concrete leaf. Best-effort: a
+    # group bundling several distinct failures is labelled by whichever leaf
+    # sorts first, not by a "mixed" bucket.
     while isinstance(exc, BaseExceptionGroup) and exc.exceptions:
         exc = exc.exceptions[0]
 
diff --git a/tests/unit/providers/test_openai.py b/tests/unit/providers/test_openai.py
index afa3cc83..00a38b0f 100644
--- a/tests/unit/providers/test_openai.py
+++ b/tests/unit/providers/test_openai.py
@@ -338,7 +338,7 @@ async def test_openai_close(mock_openai_client):
 def _req():
     import httpx
 
-    return httpx.Request("POST", "http://gw/v1/embeddings")
+    return httpx.Request("POST", "https://gw/v1/embeddings")
 
 
 @pytest.mark.unit
@@ -434,3 +434,26 @@ async def test_embed_batch_retries_on_connection_error(mock_openai_client, monke
     assert embeddings == [[0.4, 0.5, 0.6]]
     assert tokens == 7
     assert create.await_count == 2
+
+
+@pytest.mark.unit
+async def test_generate_retries_on_connection_error(mock_openai_client, monkeypatch):
+    """generate() shares the transient retry (RAG sampling survives a rollover)."""
+    from openai import APIConnectionError
+
+    from nextcloud_mcp_server.providers import _retry
+
+    monkeypatch.setattr(_retry.anyio, "sleep", AsyncMock(return_value=None))
+
+    choice = MagicMock()
+    choice.message.content = "Generated response"
+    mock_response = MagicMock()
+    mock_response.choices = [choice]
+
+    create = AsyncMock(side_effect=[APIConnectionError(request=_req()), mock_response])
+    mock_openai_client.chat.completions.create = create
+    provider = OpenAIProvider(api_key="test-key", generation_model="gpt-4o-mini")
+
+    text = await provider.generate("prompt")
+    assert text == "Generated response"
+    assert create.await_count == 2
diff --git a/tests/unit/test_processor_drop_reason.py b/tests/unit/test_processor_drop_reason.py
index 75074ebe..ee1e05c9 100644
--- a/tests/unit/test_processor_drop_reason.py
+++ b/tests/unit/test_processor_drop_reason.py
@@ -13,7 +13,7 @@ from nextcloud_mcp_server.vector import processor
 
 
 def _req() -> httpx.Request:
-    return httpx.Request("POST", "http://gw/v1/embeddings")
+    return httpx.Request("POST", "https://gw/v1/embeddings")
 
 
 @pytest.mark.unit
@@ -79,3 +79,33 @@ def test_qdrant_namespace_classified():
 @pytest.mark.unit
 def test_unknown_error_falls_back_to_other():
     assert processor._drop_reason(ValueError("nope")) == "other"
+
+
+@pytest.mark.unit
+async def test_process_document_records_drop_on_exhausted_retries(mocker):
+    """Exhausting retries in process_document increments the drop counter with
+    the classified reason (and re-raises so the outer handler counts the error)."""
+    from nextcloud_mcp_server.vector.scanner import DocumentTask
+
+    doc_task = DocumentTask(
+        user_id="alice",
+        doc_id="42",
+        doc_type="note",
+        operation="index",
+        modified_at=0,
+    )
+
+    mocker.patch.object(
+        processor,
+        "get_qdrant_client",
+        mocker.AsyncMock(return_value=mocker.MagicMock()),
+    )
+    mocker.patch.object(
+        processor, "_index_document", side_effect=httpx.ConnectError("refused")
+    )
+    rec = mocker.patch.object(processor, "record_ingest_dropped")
+
+    with pytest.raises(httpx.ConnectError):
+        await processor.process_document(doc_task, mocker.MagicMock(), max_retries=1)
+
+    rec.assert_called_once_with("connection")