Validation: can't rename to whitespace-only string.

Should return story hashes from ES results.

All-story river pages should mark returned stories unread using the all-story page.

Free users should only receive the first three river stories on page one.

Free users should be blocked before river aggregation on page two and beyond.

Newest river loads should skip stale Redis hashes that no longer exist in Mongo.

Test loading river stories normally with multiple feeds.

River pages past the reader safety limit should stop before aggregation.

Test that read_filter correctly adjusts from 'unread' to 'all' when date filter extends beyond unread cutoff.

Test loading specific story hashes without dashboard flag.

Test loading specific story hashes - THIS IS THE BUG WE FIXED. When loading specific story hashes, we should NOT run expensive ZUNIO

Test loading river stories with date filters.

Stories from the same feed should NOT form a cluster.

save(force_update=True) with no pk should not raise ValueError.

Test saving a regex author classifier via the API endpoint

Invalid regex patterns should not be saved.

Test saving a dislike URL classifier via the API endpoint

API endpoint accepts super_dislike_title and saves score=-2.

Test saving an exact URL classifier via the API endpoint

Test saving a regex URL classifier via the API endpoint

Test save_feed_chooser endpoint handles parameters correctly.

Test save_feed_chooser with approve_all parameter.

Saving a single classifier with scope=global should NOT promote other feed-level classifiers to global scope. This reproduces the bug

UserSubscription.score_story() puts super-downvoted stories in negative bucket.

search_archives should return matching archives via execute_tool.

search_archives tool should be defined.

Stories matching a super-disliked title classifier (score <= -2) must be excluded from the briefing even when their trending signals are high.

Regular dislikes (score == -1) penalize but should not hard-exclude when other signals are very strong — only super-dislikes hard-exclude.

Pairwise overlap coefficients for the reported cluster should all fall below SEMANTIC_MIN_OVERLAP_COEF. This pins the constant to the

setup_feed_for_premium_subscribers should default allow_skip_resync=False.

setup_feed_for_premium_subscribers should pass allow_skip_resync to sync_redis.

setup_feeds_for_premium_subscribers should default allow_skip_resync to False.

setup_feeds_for_premium_subscribers should pass allow_skip_resync to each feed.

Rolling Out Our New A.I. Tools' should not cluster with 'New font-rendering trick hides malicious commands from AI tools' — they share

Test loading a single feed.

Test loading a single feed with date filters.

Page 2 should extend the single-feed page cache from the existing zU key. Opening page 1 marks the subscription dirty at the end of

Staging Redis can return decoded strings for cached hashes. Keep membership checks type-normalized so page 2 does not collapse to 1 s

Single-feed unread pagination should not skip stories when read marks make the unread cache dirty between page loads.

Without feed_title_map, SMBC stories WOULD fuzzy-match (demonstrating the bug).

SMBC-style stories should NOT fuzzy-match when feed title is stripped.

soft_delete should set deleted flag and date.

Test loading starred stories.

Test loading starred stories with date filters.

Test loading starred stories with highlights.

Test loading starred stories filtered by tag.

Stats should not count deleted archives.