Changelog

Fixes confirmed problems with offline reading progress, chapter lookup recovery, download cleanup and provider retries while retaining the current site layout.

Manga

• Reading pages that have already loaded now saves progress and chapter completion even after the connection drops. Hidden tabs still do not advance reading progress.
• Chapter lookup distinguishes temporary provider failures from missing chapters, uses the existing recovery flow and validates chapter numbers before requesting them.

Downloads and services

• Generated-file cleanup waits until conversion slots are idle. An MP4 download keeps its slot through file delivery so another request cannot clean up an active download.
• Video pages set error status codes before rendering, avoiding headers-already-sent warnings when a video ID is invalid or conversion is unavailable.
• Shared service caches respect longer provider Retry-After delays and retain eligible cached results while waiting.

Maintenance

• Apache access rules cover PHP backup suffixes and editor backup files. Nginx deployments require the equivalent server rule supplied with this update.
• The cleanup upload checks now match the current supplied source, including removal of four unused interface-file requirements. Existing obsolete-file safeguards and historical release entries are retained.

Video downloads can now combine separate YouTube video and audio streams on the server instead of depending on a pre-combined format being available.

YouTube downloads

• The video download page now requests an MP4 up to 720p and uses FFmpeg to merge separate video and audio streams when YouTube makes them available to the server.
• Generation is limited to 30-minute videos, a 250 MB final file, one concurrent video job, a short per-session cooldown and three attempts per hour per visitor by default.
• Generated files are kept in the existing protected MP4 directory, streamed as downloads and removed after the request; stale leftovers are cleaned automatically.
• The downloader uses the server connection directly. Videos for which YouTube requires sign-in from the server remain unavailable rather than being routed through a visitor or proxy connection.

Quicker active-page updates and a steadier visitor list, with fewer retries during connection problems.

Online visitors

• Switching back to a tab promptly updates its active page. Busy tabs avoid queued, outdated check-ins where browser tab coordination is supported.
• Visits stay in newest-first order rather than moving on every heartbeat. Your own browser is marked You, and automatic list refreshes pause while you use a link or select text.
• Offline browsers pause requests. Failed refreshes back off, respect server retry delays and resume after reconnection; the manual refresh remains available.
• Existing counting, return-visit and privacy rules are retained. Visitor state remains outside the public web directory.

The site can use a third configured YouTube API key when the earlier credentials report a recognized limit or key failure.

YouTube reliability

• Requests try the primary key, then the existing backup, then an optional third key. Empty or duplicate keys are skipped, and a successful request stops the chain.
• Each credential retains its own cooldown. Cached results and existing shared search limits still apply; adding a key does not increase the provider’s project quota.
• Anime YouTube discovery also recognizes a third-key-only configuration. Private failure logs identify primary, backup or third without exposing credentials.

Fixes the repeated first page on Popular Anime when the catalogue provider returns the same titles for its first two offsets.

Anime

• The duplicate-boundary recovery previously limited to Airing now also applies to Popular, A–Z, Newest, categories and search/filter combinations.
• Later pages advance past the repeated provider window only after the duplicate is confirmed. Normal provider pagination keeps its original offsets.
• Recovered pagination retains the final results even when the provider’s Next link stops early. A stuck or overlapping recovery response is shown as temporarily unavailable instead of repeating the first titles.
• Existing page URLs, active filters, ten-title page size and cached requests are retained.