This is a comment left during a code review.
Path: backend/tests/external_dependency_unit/document_index/test_document_index_old.py
Line: 273-280

Comment:
**Hardcoded `doc_id_to_previous_chunk_cnt` will break future tests sharing module-scoped backends**

`doc_id_to_previous_chunk_cnt={"test_doc": 0}` is hardcoded to 0, which is only correct on the very first index call. The `document_indices` fixture is `scope="module"`, meaning the same OpenSearch and Vespa backends persist across all test functions in this module. When a second test function uses this fixture and calls `document_index.index(...)`, the backend already has 5 chunks for `"test_doc"` from the previous test run, but the batch params still claim `0` prior chunks exist. This can lead to orphaned/duplicate chunks that make subsequent assertions incorrect.

The PR description explicitly states these fixtures should be reused for additional tests, so this will become a real problem quickly. Consider making the previous chunk count dynamic or resetting it to the correct value based on backend state, or using a per-test unique `doc_id`.

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: backend/tests/external_dependency_unit/document_index/test_document_index_old.py
Line: 316

Comment:
**Fixed-duration `time.sleep` makes tests fragile in slow CI environments**

Using `time.sleep(1)` for Vespa and OpenSearch indexing propagation is a known source of test flakiness—on a loaded CI worker 1 second may not be enough for the write to be queryable. The `vespa_document_index` fixture already demonstrates a robust polling pattern (lines 145–154) that retries for up to 60 seconds. Applying a similar retry-until-consistent approach here would make these assertions more reliable:

```python
import tenacity

@tenacity.retry(
    stop=tenacity.stop_after_delay(10),
    wait=tenacity.wait_fixed(0.5),
    reraise=True,
)
def assert_chunk_count(document_index, chunk_request, filters, expected_count):
    chunks = document_index.id_based_retrieval(chunk_requests=[chunk_request], filters=filters)
    assert len(chunks) == expected_count
```

This pattern (or an equivalent simple polling loop) applies to the `time.sleep(1)` at line 342 as well.

How can I resolve this? If you propose a fix, please make it concise.