V3.10: Faster Translations, Smarter AI, Monthly Limits

Upgraded the whole translation stack to Gemini 3.1 Flash Lite Preview — the same model now powers translation, entity detection, chat, and wiki generation
New parallel chunk pipeline — the first chunk streams live while the rest translate in parallel in the background. Big chapters finish in seconds instead of minutes
Automatic failover across multiple translation accounts when one gets rate-limited — zero visible slowdown on busy days
New fallback chain for content-policy blocks: if the primary model refuses, we cascade through alternates and finally to Grok. Chapters that used to show "content omitted" placeholders now come back fully translated
Fast-fail on blocked content — we no longer waste retries on prompts the model has already refused; the next model kicks in immediately
Smoother streaming under load — the translator releases its database slot before the long LLM call, so heavy traffic no longer chokes the backend
Tuned chunk sizes and post-processing for cleaner output

Cleaner detection pipeline — entity mappings save as a baseline before the LLM pass, so even if detection hiccups mid-chapter, your chapter still has clickable entities
Junk filters added — common Chinese/Japanese/Korean words and English filler words no longer get flagged as character names
Massive glossary cleanup: removed roughly a million bloated entities that had built up from earlier over-detection runs. Glossary pages load faster and search is noticeably more relevant
Canon/fanfic auto-promotion — when a mod marks an entity edit with reasons like "canon name" or "fandom", the entity is automatically promoted to the global Fan-Fiction glossary. 91 historical entries were backfilled based on past change history
Disabling a general entity for a novel now correctly lets you add a novel-specific override with the same name
Entities with middle-dot characters (like 贞特·崔格) now match properly in detection

Wiki pages, character chat, and AI image generation are back online after the embeddings migration
Wiki and character chat now use the upgraded translation model for better quality
Dual-path chat streaming — normal chats use our fast primary path, NSFW chats route through the alternate
Character image generation switched to a faster/cheaper tier and images are now hosted on our own CDN
Wiki generation waits until a novel is at least 60% indexed before it can run — prevents half-baked wikis
Character wiki pages load dramatically faster — the main query that powered them was the biggest single source of database load and got completely rewritten (also fixed a subtle bug where character stats were bleeding across novels)

New scraper: trxs.org — that's 38 supported source sites total
Universal URL normalization — paste any URL from a supported site (novel page, chapter page, catalog, pagination, search) and the scraper now auto-resolves to the novel page. No more "invalid URL" errors because you pasted from deep in a reader
Hiatus detection fix — 92 novels that were wrongly marked Hiatus (even though they were getting new chapters) are now correctly marked Ongoing
Cleaned up a couple of novels stuck in a "0 chapters" zombie state from failed initial scrapes
Faster scraping — a new database index cut the scraper's "does this chapter already exist?" check from ~100ms down to under 1ms, which was by itself eating 64% of our database time
Inkstone uploader script improvements

Added the new chapter-URL index mentioned above — biggest single perf win of the release
"Latest updates" query on the homepage rewritten — 100× faster (went from ~1 second per call to ~11ms)
Character wiki entity query rewritten — another huge win in both speed and correctness
Novel search tuned — dropped a heavy similarity scan on full descriptions that was eating 19% of database time with almost no search-quality benefit
General FastAPI tuning across the board: fewer redundant health checks, slimmer worker pool, cleaner connection handling

Weekly request limits are now monthly. Scraping, previews, chat messages, and wiki generations now use a single monthly pool instead of expiring every Monday — unused quota carries over within the calendar month
Monthly cap = previous weekly cap × 4 (Translator Pro: 32 novel requests, 128 previews, 1,000 chats, 40 wikis per month)
Retranslations stay at 50/day — that one doesn't roll over
Pools reset on the 1st of each calendar month at 00:00 UTC
Existing weekly usage was migrated — your April activity counts against the new monthly bucket
Mid-month tier upgrades apply immediately against the new higher cap

Fixed a bug where the /me endpoint would sometimes show your level as blank and other times show the real number — now consistent on every page load
Your level badge updates as soon as the nightly XP recalculation runs, instead of staying stale for up to 5 minutes
Fixed a rare race condition in the daily rate limiters for comments, reviews, and reports
Redis cache layer now has a connection pool and faster serialization — dashboard pages load quicker
The /me/limits page cut its Redis round-trips by 80% per load
Slow-request tracking redesigned — the admin dashboard can now actually tell you which routes are slow, not just which unique URLs were slow once
Fixed a timezone bug where scraper and API services could disagree on when "this week" started, very rarely letting users double up on their quota around midnight UTC

Sitemap + RSS feed generation moved from GitHub Actions (which was getting blocked by Cloudflare) to a Celery Beat task that runs every 6 hours
The background "check for missing chapter embeddings" task now actually runs — it had been silently stalling for a while
Grok added as the final fallback for content-policy blocks
Discord mod bot rate-limit fixes and tag handling improvements

Level badges on profile pages and next to usernames across reviews, comments, community rankings, and activity feeds
New entity list sheet with plural entity handling (e.g. "Goblins" and "Goblin" are merged in display)
Custom fandom input — type your own fandom name when requesting a translation
Entity dialog in chapters opens even when the clickable-glossary toggle is off
Prefilled reasons in novel report and entity change forms
Clickable glossary hover states and styling polish
User engagement history view — new activity feed
Request view cleanup — disclaimer removed, prompts prefilled
Streaming progress visual — live progress bar while a chapter translates
Rating reminder toast after you've read enough chapters of a novel
Notification UI refresh
Chapter summary fixes, text capitalization and pluralization cleanups
Review / comment collapse in long threads
SEO optimization + repo size trim — smaller frontend bundle, faster first paint
Scrollable novel tags on desktop (not just mobile)
"Category" label renamed to "Genre" site-wide
Skeleton loading fixes across views
Back-navigation fix in the browse view
Auth flow / cookie handling fixes
Optimistic update improvements across interactions