Stats endpoint should require authentication.

Stats should return correct total counts.

Stats should correctly count today and this week.

Stats should count unique domains.

Should store content when no matching story exists.

Should store when extension content is >10% longer.

Dirty feeds still need unread reconstruction even when cached unread counts are zero. This protects against over-filtering: we want

The unread helper sometimes receives an all-subs list from higher-level callers. In that case it should only consider feeds that can

The strip regex should preserve hyphens and underscores.

The strip regex should preserve double quotes.

The strip regex should still remove other special characters.

Strip should remove special chars but keep quotes for phrases.

Stripe subscriptions with plan=None should not raise AttributeError.

Stripe subscriptions with plan=None should fall back to items.data[0].plan.

The user-facing upgrade view depends on the always_invoice default; don't regress it.

User input with quotes should have them stripped before re-wrapping.

Facebook click ID should be removed.

Feed title words should be removed from significant words.

Google click ID should be removed.

Microsoft click ID should be removed.

UTM tracking parameters should be removed.

Each subscription call increments the feed's count.

All suggested questions should be strings.

Should return at least some suggested questions.

Suggestions endpoint should require authentication.

Regression test: disabled section categories should become top_stories in LLM data.

If the hub returns a 204 response, the subscription is verified and active.

System prompt should mention archives.

System prompt should mention using tools.

System prompt should not be empty.

SchedulePremiumSetup task should default allow_skip_resync to False.

SchedulePremiumSetup task should pass allow_skip_resync to setup_feeds_for_premium_subscribers.

With odd number of quotes, last one should be escaped.

Tier 1 (title) must reject every pair — proves any cluster we see in production came from Tier 2 semantic matching, not title fuzzy.

With the tightened intersection>=4 and overlap>=0.45 thresholds, merge_clusters should NOT union these stories even if ES MLT hands th

Verify the word intersection counts that drive the threshold decision.

All tools should have name, description, and input_schema.

Stories aggregate into feeds, sorted by total read time descending.

Test that trending sites endpoint returns valid response.

Feeds are sorted by subscription count descending.

Widely Read and Long Reads should include story classifier scores.

Stories are indexed in sRTi and retrievable via get_trending_stories.

The bug: trial users have is_premium=True, the old dedupe skipped them. After the fix, activate_premium must clear is_premium_trial, send the

Unbalanced quote at start should be escaped.

Unbalanced quotes should not cause errors.

Single unbalanced quote should be escaped.

Unrecognized/missing modes should default to the merged namespace so the feature degrades gracefully when the preference is absent.

Update blocklist should add custom domains.

Update blocklist should add custom patterns.

Update blocklist should set allowed domains.

Same URL should produce same hash.

Different URLs should produce different hashes.

Normalized URLs should produce same hash.

User's allowed domains should override default blocklist.

Regression test: disabled section keys should be remapped to top_stories in curated_sections.