Test that lazy merge works correctly even for a single-feed river. This verifies there's no regression from removing the ZUNIONSTORE

Test that lazy merge works with read_filter='unread'.

If the number of seconds to lease the subscription is not specified, it should default to 864000 (10 days).

Normal news stories should still cluster when feed title is stripped.

List should respect limit and offset.

List should default to 50 items.

List should exclude soft-deleted archives by default.

List should filter by category.

List should filter by domain.

List should correctly set has_more flag.

List should include deleted when flag is set.

List should cap limit at 200.

List endpoint should require authentication.

List should only return archives for authenticated user.

List should search by title.

List response should include all expected fields.

Domain should be lowercased.

Rivian Apple Watch stories should remain clustered together.

Sales Coach story should not merge with Rivian Apple Watch stories when semantic clustering falsely matches them on shared 'apple'/'app' terms

Without story_title_map, merge_clusters should behave as before (union all).

Feeds below the minimum threshold are excluded.

Search combining phrase and regular words should work.

Moving a folder to a new parent should update folder-scoped classifier paths.

Moving a folder should also update classifiers on its subfolders.

Multiple balanced quote pairs should pass through unchanged.

Search with multiple phrases should work.

Variants list should not contain duplicates.

Stories sharing only 1-2 common words should NOT cluster.

Without feed title, all significant words are kept.

Query without quotes should pass through unchanged.

Legitimate clusters should not be affected by numeric filtering.

Pure numeric tokens like years and dates should not be significant words.

Original URL should be first in variants.