Empty phrase should return empty list without calling ES.

Empty phrase search should not crash.

Empty phrase (just quotes) should pass through unchanged.

Empty query should pass through unchanged.

If a non-202/204 status is returned, raise a URLError.

Identical titles from different feeds should cluster.

execute_tool should handle unknown tool names.

Export should return CSV file.

Export should not include deleted archives.

Export with include_content should include full content.

Export should return JSON array.

Should extract domain from URL correctly.

Should handle invalid URLs gracefully.

Regression test: h3 with class before data-section should still parse.

Regression test: h3 with style before data-section should still parse.

Test that feed autocomplete endpoint returns valid response.

Reloads should invalidate cached favicons when image data changes.

The reported bug: openrss proxy feeds were detected as YouTube feeds.

Feed-scoped classifiers only apply to stories from their own feed.

All-story river pages only need read status for returned stories. Building a full unread companion river on every all-story page cau

When feed title words don't appear in story, nothing is stripped.

Legacy username feeds should try forUsername before forHandle.

Playlist feeds should use the YouTube Data API thumbnail instead of generic favicon.

A YouTube fetch that returns nothing is recorded instead of silently dropped.

Quota exhaustion shows up as a 429 in fetch history instead of silence.

When server says ISO-8859-1 but content has UTF-8 BOM, use UTF-8.

When server says ISO-8859-1 but HTML declares UTF-8, readability should use UTF-8.

fetch_page_timeout: when server says ISO-8859-1 but HTML declares UTF-8, use UTF-8.

When server says ISO-8859-1 and HTML has no UTF-8 declaration, use ISO-8859-1.

When server says ISO-8859-1 but content has UTF-8 BOM, use UTF-8.

When server says ISO-8859-1 but HTML declares UTF-8, use UTF-8.

A feed already at the /feed/ path is not rewritten or churned.

On fetch, a legacy feed cached at the preview path rewrites itself to /feed/.

Pre-computed Redis cluster with 2+ feeds should mark ONE candidate as widely_covered.

Stories with same title but no pre-computed cluster should NOT be widely_covered.

Cluster with only 1 unique feed should NOT be widely_covered.

Pre-computed cluster with 4 feeds should mark only ONE candidate as widely_covered.

finish_fetch_archive_feeds should call sync_redis on each subscribed feed.

finish_fetch_archive_feeds should publish fetch_archive:done to Redis pubsub.

finish_fetch_archive_feeds should sleep between sync_redis calls to avoid Redis spike.

Folder-scoped classifiers are dropped when folder_feed_ids is not provided (matches existing classifier_matches_story_feed behavior) and resol

Test complete archive lifecycle: create, update, categorize, delete.