# Changelog

Important user-facing changes are documented here. This changelog is intentionally curated and high level; use the git history for commit-by-commit detail.

The format is based on [Keep a Changelog](https://keepachangelog.com/), and this project adheres to [Semantic Versioning](https://semver.org/).

## [Unreleased]

### Fixed

- Web Book resume now converts native EPUB locators through the shared canonical cursor instead of opening Foliate at the first page.
- Unified Book resume now carries forward a farther pre-migration audiobook position, preserves its within-chapter offset, and treats shared chapter boundaries as the start of the later chapter.
- Entity thumbnails now inherit canonical owning artwork where appropriate and include their visible parent title as a consistent subtitle across clients.
- Entity thumbnails now show batched, visibility-safe hierarchy and membership counts, while movies inherit technical metadata from their single playable file.
- Video playback now preserves surround channels while keeping delayed audio tracks synchronized, and safely recovers when a replaced file's stored stream map is stale.
- Direct-stream playback now preserves intentional per-track timing in fragmented MP4 without re-encoding supported surround audio.
- Entity detail pages no longer report missing-resource errors while checking optional Identify, acquisition, and monitoring state.

### What's New
- Jobs now run as durable dependency graphs with one fair interactive lane per user-selected Entity, a separately configured background pool, shared CPU balancing, durable provider rate limits, restart-safe waits, graph-level progress and cancellation, and a dependency view in Jobs.
- Prismedia now has real multi-user accounts. A first-run setup wizard creates your administrator, every household member gets their own username and password, and the same credentials sign in to the web app, Jellyfin-compatible clients (Infuse, Swiftfin, and friends), and OPDS readers — so reverse-proxy auth middleware is no longer needed in front of Prismedia.
- Everything personal is now per user: watch progress, played state, favorites, ratings, Continue Watching, and NSFW visibility belong to the signed-in account, and administrators control which libraries each member can see.
- Prismedia is now a complete self-hosted media library for videos, movies, shows, images, galleries, comics, ebooks, audio, people, studios, tags, collections, files, plugins, and background jobs.
- The app now expresses the prism in Prismedia: a neutral dark shell separates into spectrum identities for each media family, detail pages draw atmosphere from their artwork, floating chrome uses restrained frost, navigation colors are customizable, and blocking states animate light through the Prismedia mark.
- Prism color is now deliberately sparse across the web app: neutral material surfaces carry the layout while compact taxonomy markers, progress rails, selected states, and artwork-led details provide color. Browser, installed-app, and documentation icons now use the colored prism mark.
- Playback is much more capable: videos remember progress, direct-play or stream-copy when possible, transcode more efficiently when needed, expose clearer quality/status controls, and report consistent watched state to Jellyfin-compatible clients.
- Audio is now first-class, with album and artist libraries, a persistent player, queues, shuffle/repeat, waveform scrubbing, media-session controls, artwork, play counts, and restore-on-refresh behavior.
- Books and comics now include EPUB, PDF, CBZ, ZIP, and CBR reading plus first-class M4B, M4A, and MP3 audiobooks. One Book can be read, queued for listening, or opened in both formats with one shared progress position.
- Identify has been rebuilt into a durable metadata workflow with provider search, review queues, cascade matching for related children, bulk accept, Auto Identify, and safer merge behavior that preserves existing metadata.
- Request is available under Operate for finding and acquiring books, authors, movies, series, artists, and albums: Prismedia searches your indexers, downloads, monitors, and imports them itself — no external request app required.
- Library scanning is faster and smarter, with incremental rescans, batched persistence, prioritized thumbnail/metadata work, better folder organization, generated thumbnails, and safer cleanup of removed or disabled libraries.
- Jellyfin-compatible access is substantially broader, including per-user SFW/NSFW visibility, credential and session authentication, richer metadata, artwork, watched state, and compatibility fixes for clients such as Infuse, Swiftfin, Manet, Finamp, and Symfonium.
- SFW/NSFW controls now apply across browsing, plugins, identify, requests, OPDS, Jellyfin clients, thumbnails, queues, and metadata so visibility stays consistent across the app and external clients.
- Release operations are now ready for channel publishing: the root package version is the source of truth, Docker builds validate the release metadata, pushes to `main` publish only the dev image, and alpha/beta/release images are published manually.

### Changed
- Books now keep one forward-only progress position across reading and listening. Chapter-scoped mappings translate that position between text, pages, and audiobook parts, preserving exact text positions when available and falling back to the readable cursor when an audio part cannot be matched.
- Reading and audiobook heartbeats now contribute to activity time in Playback Stats, with separate reading and audiobook totals and a combined time-read-or-listened figure on Book details.
- Subtitle preference terms now accept negative weights, allowing labels such as Forced or SDH to demote matching tracks while preferred languages still promote them.
- Request and Plugins have been tightened to match the rest of the app. Plugins now state what each provider actually covers as compact per-family chips carrying that family's colour — real names like "Series" and "Audio Tracks" instead of internal codes — and the installed list uses two columns on a wide screen. Request opens on a proper content-kind chooser, sorted predictably and showing how many sources can search each kind before you pick one, and every queue row shares one artwork size, row height, and column position so titles, statuses, and actions line up down the list and across tabs. Waiting, queued, and searching rows now read as calm in-progress states instead of warnings, so a genuine failure stands out, waiting reasons wrap to a readable two lines instead of clipping mid-word, and on a phone the status and kind filters scroll in place rather than pushing the first row off screen.
- Acquisition imports now reconcile only the exact imported Entities, files, descendants, and affected ancestors. Movie, TV, music, book, retry, merged, and upgrade imports no longer trigger a whole-library or media-kind rescan.
- Identify review, metadata application, request fan-out, downloads, imports, readiness processing, and acquisition finalization remain in their originating lane; provider review and external transfers release workers while they wait and resume the same graph when signalled.
- Playback Stats has been rebuilt around the prism: one beam of playback enters the Prismedia mark and separates into a band per media family, sized by its share of the window, and selecting a band filters the whole page. Watch time, completion, library reach, streaks, and the busiest day sit beside it, an activity chart adapts from per-day to per-week columns to stay readable at any width, and a weekday-by-hour rhythm grid shows exactly when the library gets used. Charts now use the spectrum as a cool-to-warm scale so hue reads as volume, and every list shows real artwork, how far playback actually reached, and accumulated watch time.
- Delete files now performs a complete removal: the selected media subtree, managed files, generated assets, monitors, acquisitions, and stuck lifecycle state are removed together instead of monitored content returning to Wanted.
- Browse, Settings, navigation, ambient backgrounds, and missing artwork now share one restrained rainbow palette in deliberate spectrum order across the web and Apple apps.

### Added
- Acquisition profiles can now defer automatic searches until a provider-supplied release milestone (such as digital, streaming, or physical availability), with an optional delay; waiting requests stay monitored, show their release state, and appear with every known milestone on the new release calendar. Profiles search immediately by default, request and identify metadata apply release dates up front, and a missing milestone gets one uncached provider refresh before users can wait for metadata, search manually, or enter a named release date in the edit pane.
- Metadata plugins can now provide typed semantic dates, including theatrical, digital, streaming, physical, broadcast, and publication milestones, while legacy date keys remain compatible and normalize at ingestion.
- Collections now belong to their creator and stay private by default, with an edit-screen sharing control for household visibility. Shared collections remain owner-managed, and dynamic rules evaluate play and skip activity using the owner's personal history.
- Acquisition file results now retain a privacy-safe per-file import record, including source-to-library mapping and clear imported, skipped, or failed outcomes, so completed and partial imports remain understandable after the job finishes.
- Added first-class Soulseek support through slskd: Prismedia can search peer results as album folders or individual tracks, show their remote folder context with a dedicated Soulseek chip in release review, queue them as atomic slskd batches, monitor and import their files, and configure Soulseek alongside torrent and Usenet clients. Album requests now materialize their track list and automatically fall back to artist/album/track-qualified searches for anything still missing after a failed or partial album acquisition, while searches allow a full peer-response window instead of ending after only a few milliseconds.
- Acquisition review now supports deliberate replacement and local content: on-disk books, movies, episodes, and albums can run an ephemeral replacement search that never auto-selects a release, every release picker accepts an exact custom search term, and manual file or multi-file uploads show transfer progress before joining the same import or in-place upgrade workflow as a completed download. Album replacements stage the complete multi-file release before swapping it into the existing library folder. ZIP uploads are inspected and safely expanded when they wrap media, while comic archives remain intact.
- Blocklisted acquisition releases are now manageable at scale: Request History can allow a blocklisted release again directly from its event, while Settings groups releases by their associated work, searches across work, release, indexer, hash, and reason, and incrementally reveals large result sets.
- Blocklisted acquisition releases can now be cleared in bulk for one library item or across all items, with recent-time ranges for quickly recovering from a temporary download-client outage.
- Series and seasons now track their own episode progress: partially watched episodes resume in place, completed episodes offer the next one as Continue, and starting in a later season advances the series without marking earlier seasons watched. Both detail pages use the shared progress card with independent watched and start-over controls.
- Files can now be downloaded directly from the file browser. Downloading a folder prepares a ZIP in the background, shows live file-count and compression progress, and starts the browser download automatically when the archive is ready.
- Books with both readable and audio editions now have one combined progress card for continuing either rendition or opening both together. All three actions resume the same farthest saved position, estimate the matching page or timestamp within its chapter, and start audio five seconds early for easier manual alignment.
- Book detail pages now turn EPUB contents and multipart audiobooks into one chapter list with artwork-colored reading and listening markers, per-chapter Read and Listen actions, and a combined mode that opens the matching text and audio together.
- Added first-class OpenSubtitles acquisition: configure and test an account in Settings, find explainable hash/identity-ranked matches from a video's Transcript tab, import a selected subtitle into app-owned playback assets, or automatically fill missing configured languages using a strict confidence threshold.
- Acquisition searches now label Usenet and torrent releases, prefer the configured download type before falling back, filter results to enabled download-client capabilities, and let Identify, Request, and release review reveal more than ten matches.
- Entity library queries can now request several kinds at once, allowing native clients to load one server-filtered movie and TV activity feed without duplicate movie file rows.
- Video libraries now discover subtitle files beside their videos automatically. Same-name SRT, VTT, ASS, and SSA files — including language or descriptive filename tags — are normalized into app-owned playback assets, and rescans reconcile additions, edits, renames, and removals.
- Added a first-run setup wizard: a fresh install (or an upgraded one, on its next visit) walks through creating the administrator account, and an upgraded install also reviews the accounts migrated from its former Jellyfin sign-in profiles — each keeps the previous server API key as its password so connected apps keep working, with an inline option to set a new password on the spot.
- Added a login page and signed-in sessions for the web app: sign in once per browser and stay signed in (sessions slide forward for 90 days of use), with a user chip at the bottom of the navigation for Account, Manage users, and Sign out.
- Added user management under Settings → Users: create, edit, disable, and delete accounts, reset passwords, grant or revoke NSFW visibility, allow specific members to create their own libraries, and pick exactly which libraries each member can see (admins always see everything). The last enabled administrator is protected from demotion, disabling, and deletion.
- Added per-library access: new libraries are visible only to their creator until shared, admins can grant access per member from the library list, and members holding the create-libraries permission manage the libraries they created from the same Settings surface.
- Added per-user watch state: progress, played state, play counts, favorites, ratings, and reading positions are tracked per account across the web app, Jellyfin-compatible clients, and OPDS. Upgrading copies the household's existing history to every migrated account, so nothing is lost.
- Added an Account page for every user: update your display name, change your password, choose your content visibility (when permitted), and review or sign out connected devices and sessions.
- Added a host-level recovery escape hatch: set the `PRISMEDIA_RECOVERY_PASSWORD` (and optionally `PRISMEDIA_RECOVERY_USERNAME`) environment variables to reset or create an enabled administrator on boot, then unset them.
- Added a durable acquisition history: every grab, import, import failure, download failure, blocklisting, upgrade, and removal is recorded in a permanent activity log that outlives the acquisition it describes — deleting or cancelling a request no longer erases the record of what happened. A wanted item's page shows its own history inline, and Request gains a global History tab listing all activity newest-first with per-event and per-kind badges, the release, indexer, and download client involved.
- Added custom formats — named, scored release classifiers in the Sonarr style: define a format from conditions over a release's title (regex), release group (regex), language, or quality, then give it a score in each profile. Matching formats add their score to a release's ranking (100 points equals one preferred term, and negatives act as a soft ban), a "minimum format score" rejects anything below the floor with a visible reason, and a "cutoff format score" keeps a monitored movie or episode upgrading toward a better-scoring release even at the same quality. The classic case — a format matching English/dual-audio scoring +500 to flip which release wins — now works across books, movies, TV, and music.
- Releases now understand PROPER/REPACK revisions: a proper or repack outranks the plain release at the same quality, and — under the new "Download propers and repacks" setting (prefer and upgrade / do not upgrade / do not prefer) — a same-quality higher revision of an owned movie or episode counts as an upgrade and is swapped in automatically. Anime-style v2/v3 version tokens are understood too, and release groups are now extracted from titles as groundwork for custom format scoring.
- Movie, TV, and music profiles now speak a real quality ladder: video releases place on a combined source × resolution ladder (SDTV up through Remux 2160p) and music on a codec ladder (lossy up through hi-res lossless). A profile can allow exactly the qualities you want — pinning tiers or capping the ladder, the "never chase 4K" setup — and carries an upgrade-cutoff quality for the upgrade loop. Release ranking uses the ladder position instead of loose token heuristics, and disallowed qualities are rejected with a visible reason.
- Added an opt-in recycle bin for upgraded-away files: set a recycle-bin folder in Settings and a file replaced by a quality upgrade moves there (in dated subfolders) instead of lingering beside the new one as a .prismedia-bak sidecar; a daily cleanup purges entries older than the configurable window. Leave the folder blank to keep the previous sidecar behavior.
- Torrents now respect seed goals instead of being deleted the moment they import: set a seed-ratio and/or seed-time goal per indexer (with per-client defaults for grabs whose indexer sets none), and a hardlink- or copy-imported torrent keeps seeding until either goal is met — then Prismedia removes it and its download-dir data automatically, leaving the library file untouched. Goal-less torrents are left to the client's own rules, and private-tracker ratios stop being collateral damage of an import.
- Added hardlink imports: a new "Hardlink (instant, keeps seeding)" import mode places the library file as a hard link to the still-seeding download — instant, no extra disk space, and the torrent keeps its ratio — falling back to a copy across filesystems. Movie, TV, and music imports now also honor the profile's import mode (previously always move) and leave the torrent seeding for hardlink/copy imports.
- Added remote path mappings for split-host and containerized download clients: map a client's path prefix (its own filesystem view, e.g. /downloads) to where those files are visible to Prismedia, per client, under Settings → Acquisition. Longest matching prefix wins, Windows-style client paths translate, and deployments on one filesystem need no mappings at all.
- Downloads carrying an executable or dangerous file (.exe, .scr, .bat, .lnk, and friends — the classic fake-release payload) are now held for review instead of imported: the acquisition shows a clear warning naming the file, nothing touches the library, and the release picker reopens so you can block the release and grab a different one. Releases whose title already names such a file are now also rejected before they are ever downloaded, closing the obvious abuse loop up front.
- Request discovery now starts with a content kind and metadata source, then shows the exact search fields that plugin supports — such as series title and year or book title and author — while preserving the plugin's own result ranking.
- Indexers are now reliability-managed: a failing indexer backs off on an escalating ladder (1 minute up to 24 hours) instead of polluting every search with the same error, recovers gradually as searches succeed, and shows its back-off state in Settings. Indexer priority now genuinely matters — equal-quality releases prefer the lower-priority-number indexer — and each indexer can carry an hourly query budget that searches respect.
- Prismedia now talks to indexers directly — no Prowlarr required: add any Torznab (torrent) or Newznab (usenet) indexer by URL and API key, including Jackett's per-indexer endpoints and native tracker APIs. Prismedia reads each indexer's capabilities (advertised categories and page-size limits, cached for a week) and trims searches to what the indexer actually supports. Prowlarr remains fully supported as an aggregator alongside direct indexers.
- Added Transmission as a second torrent client, and download clients now route intelligently: each release goes to an enabled client that speaks its protocol, clients carry a priority (lower wins) so you control which is preferred, and when the preferred client refuses an add Prismedia falls back to the next one instead of failing the grab. Acquisition profiles gain a per-kind download category so movies, music, TV, and books can stay segregated inside the client (blank keeps the client's own category).
- Added usenet support: configure SABnzbd as a download client and Prismedia acquires NZB releases end to end — usenet releases from your indexers are accepted and ranked alongside torrents, sent to SABnzbd under your category, tracked through download and unpack, and imported on completion. Each release is routed to a client that speaks its protocol automatically, and a download SABnzbd reports as failed (incomplete, unpack error, encrypted) goes straight to failed-download recovery instead of waiting out the stall window.
- Requesting now creates the item in your library immediately: committing a request builds the book — or the author with each selected book — as a real library entity marked Wanted, populated with the provider's metadata and artwork before anything downloads. The download that fulfils it attaches to that same entity, so nothing appears twice. Re-requesting something you already own or already have in flight is recognized and skipped transparently.
- A wanted book is managed from its own page in the library: the book's detail page now carries the full acquisition surface inline — release picking, live download progress, Search again, Monitor, and Cancel — so wanted items and real items live and are managed in the same place. Cancelling from there removes the placeholder and returns you to the library.
- Wanted items can be removed in bulk: select placeholders in any grid and "Remove wanted" deletes them (tearing down any in-flight download) and keeps them out of future discovery — a followed author/artist won't re-add a work you removed. Requesting that exact item again later brings it back and clears the block.
- Release searches now drill down with context: each search phrase is built from the item's lineage (an album leads with its artist, a movie with its year, a book with its author) and falls back to broader phrasings only when nothing acceptable is found — the same top-down rule TV will use (whole series, then season packs, then single episodes).
- The release picker gains a "Show only relevant" toggle (on by default) that hides rejected releases — wrong season/episode, below minimum seeders, and the like — so the acceptable grabs stand alone; turn it off to review every candidate the search returned.
- Monitored items can be checked on demand: a "Check monitored items" job in the Jobs switchboard runs the whole sweep now, and a followed author/artist page gains "Check for new works" to re-sync just that one immediately.
- Request gains Wanted lists in the Sonarr style: a "Missing" tab lists everything monitored but not yet acquired, and a "Cutoff unmet" tab lists items you own that are still below their quality cutoff (showing owned → cutoff quality). Both are paged and filterable by kind, show each item's last-searched time, next-search ETA, and stuck-search count, link to the item's library page, and let you select rows to Search now or Unmonitor in bulk.
- Requesting is now hands-off: a requested item automatically searches, picks, and downloads the best acceptable release — the release picker only appears when nothing acceptable was found — and stays monitored on a daily schedule until it's acquired or you turn monitoring off.
- Authors and artists can be followed for new works — requested ones are followed automatically, and any on-disk author/artist can be monitored from its own page once Identify has given it a provider identity. The daily sweep sends newly discovered books and albums through the same monitored child-acquisition path as a direct toggle: each appears as a fully described Wanted Entity and searches or downloads according to its profile. Turning an individual child off remains authoritative even while its parent is followed, so the next sweep cannot silently recreate it.
- Discover/Request now covers movies and music alongside books: search TMDB for a movie or MusicBrainz for an artist, browse an artist's discography, and request — a wanted Movie or Album entity appears in your library with full metadata, cast, and artwork, and its release search grades movie releases by resolution/source and music by codec quality (lossless first), querying your indexers under the right categories per kind. A wanted movie's page shows the same inline acquisition surface books have, and movie/music imports complete through the same monitored download-to-library loop.
- TV series complete the request lineup: search a series, pick the seasons to get, and each becomes its own season-pack acquisition searched top-down with series context ("Show S01", "Season 1", complete-series packs) while every episode of a picked season appears as a clearly-badged Wanted placeholder under its season. A finished download imports as `Series/Season NN/Series - SxxEyy.ext` — the exact layout scanning expects — and immediately binds the files to the series, season, and episodes the request created, matching episodes by position with no duplicates. Episodes the pack didn't include stay Wanted; open one and press "Search for release" to fetch it individually (its search is built from the series and episode numbers it carries). Series pages gain the same Follow/Check-for-new-works monitoring authors and artists have, wrong-season or single-episode releases are rejected when a pack is wanted, and a new TV quality profile kind governs it all.
- Requesting a series, author, or artist now offers the same medium-neutral monitor policies: All current and future, Missing now, Future only, or Manual selection. The shared child picker handles seasons, books, albums, and future Entity hierarchies without a TV-only Season Pass model. All and Future send newly discovered direct children through the normal monitored acquisition path, while turning a child off records an explicit suppression so its parent cannot add it back.
- Parent Entity pages now use one shared child-monitoring surface: series manage seasons, seasons manage episodes, authors manage books, and artists manage albums through the same per-child controls and bulk Monitor all / Unmonitor all actions.
- Search results only come from providers that can also open their details, so a result can no longer dead-end on "Request detail was not found" (and the YouTube plugin learned to resolve albums and artists by id, fixing exactly that for YouTube Music results).
- Movie and music downloads now import themselves, completing the hands-off loop for those kinds: a finished movie download places its main video as `Title (Year)/Title (Year).ext` in your first video-enabled library (skipping sample files, and stopping for manual review when a release holds more than one full-size video), and a finished album places its tracks — keeping disc folders, carrying cover art — under `Artist/Album/`. Import immediately binds those files to the exact wanted Movie, Album, and Artist Entities the request created, so the content is ready without waiting for a scan, nothing appears twice, the torrent is cleaned up after the move, and the item's monitor marks itself fulfilled.
- Added an Authors library: books organized as `Author/Title` are grouped under a browsable author, mirroring how artists group albums — selecting an author shows their books.
- Added OPDS 1.2 catalog support for external ebook and comic readers, including authenticated feeds, search, covers, downloads, profile visibility rules, and reverse-proxy-aware links.
- Added playback history and a Stats page for plays, skips, top items, recent activity, and timeframe-based summaries across video, books, and audio.
- Added first-party book acquisition: request a book and Prismedia searches your indexers through Prowlarr, scores releases, downloads the chosen one with qBittorrent, and imports it into your library where it is auto-identified. Book and author search come from the OpenLibrary plugin in the same Request view, with configurable indexer/download-client/profile settings, live transfer progress with a piece map, an imported-files view, and — for releases an indexer only links to a web page — automatic magnet resolution from that page plus a manual .torrent upload fallback.
- Added failed-download recovery for book acquisition: when a healthy download client reports a download failed, or it stalls with no usable peers for an extended period, Prismedia blocklists that release so it is never grabbed again, and — when the new per-profile "auto-redownload" option is enabled (off by default) — automatically grabs the next-best acceptable release with no manual step. A brief stall is tolerated, so only a sustained one is treated as a dead release. Acquisition settings now expose the auto-grab and auto-redownload toggles and a blocklist panel for reviewing and removing blocklisted releases, and you can block an unwanted release directly from a request's release list.
- Added rich request-time metadata for book requests: the cover, description, and dates are captured and held the moment you request a book — so the request page shows real artwork and a synopsis instead of a blank placeholder — and are used to pre-fill the imported book. When the quick search result is missing a cover or description, Prismedia enriches it in the background from the provider by work id. The authoritative metadata pass (and any children) still runs at import via identify; the held data only fills gaps and never overwrites it.
- Added a "Search again" action on a book request so you can re-run its release search on demand (the manual counterpart to monitoring), without creating a duplicate request.
- Request review now uses the same proposal-review building blocks as Identify for every supported medium: choose seasons, books, volumes, or albums directly from the provider's real structural proposal, while standalone movies, episodes, books, and albums request their reviewed root item.
- Added release-selection rules to book acquisition profiles: preferred, required, and ignored terms are now editable, and preferred terms (e.g. "retail", "epub") rank a matching release above others so the best acceptable copy is chosen automatically rather than just the most-seeded one.
- Requesting now asks instead of assuming: every request page carries one request card holding the quality profile, the library to import into, and the Request button together. The profile leads — switching it re-targets the library to that profile's default, which you can still override — and the options are filtered to the media kind and the current SFW/NSFW view. The choices ride the whole pipeline: the release search uses the chosen profile's rules and the import lands in the chosen library, a followed author/artist remembers them so newly discovered works inherit the same choices, and a stale choice (deleted profile, disabled library) degrades safely to the defaults instead of failing.
- Request candidates stay attributed to the selected plugin and open a canonical proposal review instead of impersonating library Entities before they exist.
- Request review shows the canonical metadata, artwork, structural children, and related records that Identify would apply, and commits are bound to that exact plugin and proposal revision so a changed result must be reviewed again before anything downloads.
- Acquisition profiles are now per media kind: create separate Books, Movies, and Music profiles — each with its own default, target library, seeders/size/term/language rules, and weighted terms — and the right one governs each kind's searches and imports automatically. Book-only settings (path template, format cutoffs, upgrade-until-cutoff) stay book-scoped; movie and album placement is fixed to match library scanning (`Title (Year)/Title (Year).ext`, `Artist/Album/`).
- Added custom weighted terms and preferred languages to release selection, applied across every media kind: weighted terms ("remux: 150", "upscale: -200") add or subtract their weight from a matching release's ranking so you can tune exactly what wins, and an ordered preferred-language list (defaulting to English) skips releases tagged only with other languages — recognizing common title tags like FRENCH, GERMAN, ITA, and MULTi — while ranking earlier-listed languages higher. Untagged releases count as your first language, and multi-language releases always qualify.
- Added monitoring for book requests so a wanted book is no longer one-and-done: mark it monitored and Prismedia keeps re-running its release search on a schedule until an acceptable release appears (and, with auto-grab on, downloads it). Monitors reconcile themselves — fulfilled once the book is acquired, paused if you cancel or remove it — and the re-search cadence is configurable in settings.
- Added "upgrade until cutoff" to book acquisition profiles: once a monitored book is imported, Prismedia keeps watching for a higher-quality release (ranked by source — retail over web — and format) and, when one that beats your copy without downgrading it appears, downloads and swaps it in automatically, stopping once the configured cutoff quality is reached. The swap is conservative and safe: the upgrade downloads alongside your library without touching the owned file until the new copy is in hand, the replacement is in-place with the previous file kept as a recoverable backup, a release can never downgrade your format, an untrusted "retail" tag alone never overwrites a file, and a format change (e.g. PDF→EPUB) is surfaced for manual replacement rather than applied automatically. If no better release ever appears it gives up after a bounded, backing-off number of tries. Requires auto-grab.
- Added persistent audio playback with a global player bar, mini-player, queue drawer, artist queues, media-session integration, waveform rendering, track ratings, and saved local playback state.
- Added first-class audiobook renditions to Books: book libraries scan single-file and multipart audiobooks, preserve ebook and audiobook editions together, resume listening across audio parts, and let Request acquire an audiobook onto a new or existing Book through audiobook-specific indexer categories.
- Added EPUB and PDF scanning plus built-in readers with paged or continuous layouts, zoom, search, table of contents/outline support, links, downloads, and saved reading progress.
- Added configurable sidebar and mobile navigation sections, with rename, reorder, hide, collapse, reset, pinned mobile destinations, and server-saved layout preferences.
- Added dynamic and hybrid collections, including rule previews, media-specific filters, library-root filters, watched/progress filters, and collection cover modes.
- Added audio-capable collections that can act like playlists when they include artists, albums, or tracks, including a reusable Audio tab and Jellyfin playlist exposure for compatible clients.
- Added track-level multi-select to audio track lists, including select-all, Add to Collection, and shared bulk action controls.
- Added automatic hourly collection refresh, enabled by default, so dynamic collection rules stay current without a manual refresh.
- Added durable Identify queues, bulk identify, provider seeking, auto identify settings, cascade child matching, live apply progress, and review screens for movies, shows, seasons, episodes, books, music, galleries, people, studios, and tags.
- Added first-class editing for metadata, credits, roles, characters, studios, tags, links, dates, provider IDs, ratings, classification, poster/header artwork, and related entities across detail pages.
- Added operational controls for background jobs, worker status, transcode cache limits, file uploads from the Files tab, reversible scan exclusions, and centralized app settings.
- Movie, TV, and music profiles now carry a configurable naming template, the same way books do: choose how imported files are named — for example a movie as `{Title} [{Quality}]/{Title}.{ext}` or a TV episode as `{Series}/S{Season:00}/{Series} {Season:00}x{Episode:00}.{ext}` — using tokens like `{Title}`, `{Year}`, `{Series}`, `{Season:00}`, `{Episode:00}`, `{Artist}`, `{Album}`, and `{Quality}`. Each kind keeps a sensible default (leaving the field blank restores it), the folder structure stays constrained to what library scanning binds against, and an invalid template is refused at save with a clear message.
- Downloads are now validated against their actual file lists: as soon as the torrent's metadata names its files, a grab whose contents contradict what was searched for (a different year's remake, the wrong season, or a payload missing the sought episode) is abandoned, blocklisted, and replaced with the next-best release automatically — and at import time the same check holds a wrong-looking payload for review instead of renaming it into the expected title's folder. Releases you pick or upload yourself are never second-guessed.
- Media library items can now be deleted, files and all, from one shared Entity action across movies, series, seasons, videos, galleries, images, books, volumes, authors, albums, tracks, and artists. Deleting is monitor-aware — a directly monitored Entity goes back to Wanted and starts a clean replacement search, while an unmonitored Entity is removed from the library entirely. Turning off a child remains authoritative even when its parent is followed, and unsafe archive-member chapters/pages never expose file deletion.
- Cancelling or removing a download is no longer a give-up: cancel stops that transfer and removing a download from the Downloads view cleans up the record, but in both cases the item stays Wanted and its monitor stays active — the loop searches again on its normal cadence (blocklist the release from the picker if it should never be grabbed again). The explicit give-up paths remain "Remove wanted" and unmonitoring.
- Monitored series, seasons, and other containers now offer "Search missing content" on their Acquisition tab: one click sweeps the whole item at any depth — including episode gaps hiding inside partially-downloaded seasons — and starts a monitored search for every gap, reporting how many it covered. The existing "Search N missing" action remains when gaps are already visible.

### Changed
- Manual channel publishing now cascades through less-stable channels: release refreshes release, beta, and alpha images; beta refreshes beta and alpha; and alpha remains alpha-only.
- Identify and Request can now start with a different preferred metadata provider for every Entity kind, while safely falling back to the first compatible enabled provider when a saved choice is unavailable.
- Acquisition tabs now begin with a compact Monitor toggle and reveal search, activity, child monitoring, and other acquisition controls only while monitoring is enabled.
- Automatic subtitle selection now uses a case-insensitive weighted term list: every matching language or label term contributes to a track's score, making preferences such as English + Forced additive while preserving compatibility with existing language lists.
- Detail-page provider chips now show only the provider name, keeping long opaque IDs out of the layout while preserving the full linked identity as accessible context.
- Identify and Request proposal reviews now use compact media-shaped thumbnails, make the full card useful, support Select all / Deselect all, and let container results drill into their held child details without repeating provider work.
- Identify and Request reuse the reviewed proposal when accepting it, stage required artwork concurrently, and can progressively load up to 100 provider search results when the provider supports it.
- Music navigation now separates Albums, Artists, and Tracks, with a library-wide track list that can play or shuffle every song while preserving each track's album artwork and artist context in Now Playing.
- The web now-playing chrome now derives its outline, active transport controls, progress rail, and waveform accents from each current track's album artwork while keeping its ground surface opaque and neutral.
- Popovers, menus, command search, confirmation prompts, and other transient controls now share a restrained neutral-glass treatment, with page color reserved for compact accents such as the sidebar section chevrons.
- Rejected acquisition releases can now be selected manually when the rejection is an overridable match or profile rule; unsafe, blocklisted, and otherwise unqueueable releases remain protected.
- Identify and Request now use the same provider search surface, so a fresh Identify search exposes every field declared by the selected plugin instead of falling back to a title-only query.
- Desktop EntityGrids now default to a balanced mid-range thumbnail size, while preserving any saved per-grid preference.
- Library grids now filter by shared Availability states instead of the ambiguous Has file toggle: On disk uses real source-media ownership, Wanted uses request state, and every acquisition stage from Pending through Imported, Failed, Cancelled, and Needs attention is available consistently on library and detail-page grids.
- Stored external identity namespaces are normalized during upgrade and protected by database constraints and a provider/value lookup index, while intentionally allowing the same upstream item to represent more than one local copy.
- Identify metadata application now resolves and persists all provider identities through the canonical Entity identity store, so multiple IDs are evaluated together and transient URL locators cannot leak into durable identity data.
- Requests now resolve provider identities through the same canonical entity identity system as metadata imports; ambiguous duplicate identities stop with a clear conflict instead of silently binding to an arbitrary library item.
- External identity lookup now returns every matching local Entity and reports ambiguity explicitly; identity writes use named add, update, or replace policies without silently committing another operation's transaction.
- Plugin protocol compatibility is now enforced for both registry and installed manifests, and every plugin request uses one canonical protocol version instead of scattered literals.
- Plugins can now declare the external identity namespaces and schema-driven search fields they support, while legacy manifests receive a compatible title search and plugin-id identity namespace automatically.
- Entity details now show the authoritative metadata and monitoring plugin with its persistent provider ID; when that plugin declares a provider-page URL format, the shared chip opens the exact movie, series, season, episode, person, or studio in a new tab. On small screens the provider and other identity chips move to their own wrapping metadata row instead of clipping the hero actions.
- Identify search forms now come from the selected plugin's declared field schema, so context such as year or series title is sent instead of being displayed and silently ignored.
- Monitoring now targets the Entity's stable plugin identity rather than treating an acquisition row as the durable intent. Turning monitoring off removes pending, review, failed, and download state while preserving verified on-disk files; turning it back on reuses the same Entity target across books, albums, movies, seasons, episodes, and future media kinds.
- Monitoring, request lookup, and metadata enrichment now route Entity identities through plugin-declared namespaces, so a plugin's installation ID no longer has to match the upstream ID namespace it handles.
- External provider identities now use one validated, normalized namespace/value model across the Entity domain, laying the foundation for plugin-independent identify, discovery, and monitoring dispatch.
- A download held for manual import is now actionable in place: an "Import anyway" button re-runs the import with your consent to replace the existing file even when the format differs (e.g. an mkv upgrading an mp4) — the previous file moves to the recycle bin when configured (or stays as a recoverable backup), quality gating still applies (a format change never lets a downgrade through), and the dangerous-file hold is never bypassed.
- The acquisition card gains a "Search N missing" action wherever an entity still has unrequested wanted children — a season with absent episodes, a series with unrequested seasons — and "Search again" is now also offered while an item is held for manual import.
- The acquisition card's imported file list now collapses by default (matching the History section) so a large season pack no longer fills the page; it stays expanded while a download is still in flight, and expanding or collapsing it by hand now sticks instead of being undone by the next live refresh.
- Library grids now load dramatically faster: every entity with artwork gets small grid-sized cover variants (standard and high-DPI) generated automatically — including movies, series, books, albums, people, and studios whose posters come from metadata providers — and grids download those instead of the original full-resolution artwork. Existing libraries heal themselves via a background sweep on upgrade, and browsers now cache artwork properly instead of re-fetching every image on each visit.
- Generated video thumbnails now follow the thumbnail-quality preset sensibly: each preset caps the image at a reasonable resolution (e.g. High = 1280px wide) instead of producing near-source-resolution, near-lossless files.
- Settings now opens as a section directory, with each settings area on its own focused page and contextual links landing directly in the right section.
- Upgrading brings your existing setup into the account era automatically: former Jellyfin sign-in profiles become real user accounts with access to every existing library, their password is the previous server API key so Infuse, OPDS readers, and other connected apps keep working unchanged, and already-signed-in client sessions stay signed in across the upgrade.
- NSFW visibility is now a per-user permission. The browser's SFW/NSFW toggle works only within what the signed-in account is allowed to see, and Jellyfin and OPDS clients follow the same per-account rule — network position no longer plays a part.
- Server administration (settings, files, jobs, identify, requests, acquisition, plugins, users, and system operations) now requires an administrator account; members browse, play, and manage their own profile, favorites, and ratings.
- Request gains a global Downloads tab: every active download across books, movies, TV, and music in one place, with live progress, speed, ETA, size, and the download client carrying each transfer. Rows link to the item's own library page and support single or bulk Remove and re-search. It replaces the previous card-based Requests queue (finished and cancelled items live in History).
- Redesigned the Request list tabs (Downloads, Missing, Cutoff unmet) as an engaging shared card list instead of a dense table. Each row shows the item's real cover art (in its natural shape) beside its title and creator, a clear status chip with a plain-language line ("Select a release to start the download.", "Waiting for a download slot"), and — while downloading — a progress bar with the download client, speed, size, and ETA. Every row carries a primary action matched to its state (Choose release / Search again / View), Remove, and an overflow menu, and the lists gain a text filter, status and kind chips, and a sort control. All three tabs share one design and restack cleanly on mobile, so the look stays consistent and improves everywhere at once.
- Acquisition and monitoring now live in a dedicated Acquisition tab on the item's own page: the entity detail tabs gain an Acquisition tab — appearing only while the item actually has an acquisition story — combining "Search for release" (for wanted placeholders), the follow/monitor toggle with "Check for new works" (for authors, artists, and series), and the full download management surface (release picker, live transfer, files, cancel). The separate acquisition progress page under Request is gone — everything routes to the item's real library page, and requesting a single item lands you directly on it.
- A series, season, author, or artist page now rolls up its children's acquisition status in its Acquisition tab: each wanted season/episode/book/album is listed with what it's doing (Failed, Searching, Review…), so per-child progress reads from the container without opening each one.
- Following an author, artist, or series for new works is now offered only when an installed metadata plugin can actually track it — the entity must carry a provider identity (from a request or from Identify) that an enabled plugin can re-resolve by id. The card names which provider the watch rides on, and the server enforces the same rule.
- Movies and single TV episodes now upgrade until their profile's cutoff quality, the same way books do: an imported movie or episode stays monitored and re-searches for a higher-quality release (graded on the video ladder), and when a strictly-better one is found it is atomically swapped in for the owned file — the original is kept as a recoverable backup (or sent to the recycle bin) — until the cutoff quality is reached. Turning the profile's upgrade-until-cutoff toggle off, or reaching the cutoff, fulfills on import. Season packs and albums capture their owned quality for display but fulfill on import (a multi-file pack isn't swapped in place). A format change (e.g. mkv → mp4) is left for manual replacement, exactly as with books.
- First-party acquisition now replaces the older upstream Radarr/Sonarr/Lidarr request-service integration. Existing legacy request-service and request-history database tables are retained as unmanaged data during migration instead of being dropped.
- Book and author discovery opens a detail review page before requesting: clicking a result no longer immediately queues it. You review the cover and synopsis, multi-select which books (or series volumes) to get, then explicitly Request — at which point Prismedia searches and downloads the selected book(s).
- Reworked the dashboard around Continue, Recent, recently added media, and a resume-focused hero so active items appear immediately instead of being buried in per-type shelves.
- Standardized detail pages and library grids on shared Prismedia building blocks, improving filters, sort, pagination, thumbnails, media-wall/feed views, responsive controls, and cross-page consistency.
- Made desktop library grids and Identify review thumbnails open denser by default so large screens show more items without oversized cards.
- Made scans incremental, batched, deduplicated per media kind, and priority-aware so large unchanged libraries finish quickly while new items surface with useful metadata and covers first.
- Improved media organization rules for loose videos, movies, direct-child series, audio folders, ebook/PDF folders, galleries, single-image folders, comic archives, sidecars, and generated assets.
- Reworked video playback negotiation around client capabilities, direct play, direct stream, stream-copy, hardware-aware transcoding, clearer manual quality selection, and lower CPU usage.
- Rebuilt Identify around durable server jobs instead of fragile in-memory page state, with user-triggered searches taking priority over background auto-identify work.
- Expanded Jellyfin-compatible responses with richer Prismedia metadata, artwork, media sources, playback progress, collections, people filmographies, home shelves, and profile-scoped libraries.
- Tightened API and deployment behavior so every `/api/*` route requires a signed-in user, the web app carries its session cookie same-origin automatically, and native and third-party clients authenticate with the same bearer tokens the login endpoint issues.
- Refreshed release workflows so validation, documentation publishing, and alpha/beta/release channel images are manual while `main` publishes only the dev image.
- Updated documentation, README screenshots, branding assets, install metadata, and app copy to match the current Prismedia v1 surface.

### Fixed
- Jobs now separate dormant signal waits from active execution lanes, making it explicit that review and external-event waits preserve workflow state without consuming worker capacity.
- Unknown embedded subtitle streams are now skipped, and subtitle extraction remains best-effort so unsupported tracks cannot fail an otherwise healthy library scan.
- Interactive acquisition imports now resume immediately after download completion, including imports already queued before upgrading; graph continuations declare their entity gate automatically and startup repairs missing scheduler state.
- Large scan graphs no longer fail media probes, subtitle extraction, or previews through graph-level PostgreSQL concurrency conflicts, and stale graphs whose nodes are already terminal repair their displayed status automatically.
- Audio probes now persist complete media-source state, preventing unchanged tracks from being re-probed on every scheduled audio scan.
- Jobs now distinguish completed and failed lanes from genuine waits, count all terminal nodes in progress, expose failed-node totals, and describe entity serialization as queued exclusive access instead of an unexplained change wait.
- Streaming release gates now accept provider-supplied digital/VOD dates, remain visibly Waiting for release until the resolved date, and keep manual search as an action instead of a misleading lifecycle state. The date-entry prompt appears only after a completed provider lookup returns no compatible milestone.
- Release calendar entries for seasons now include their series title and link to the season detail, while crowded dates collapse into an accessible full-day list instead of overflowing the month grid.
- Durable provider gates now persist PostgreSQL leases and start intervals atomically, so concurrent workers cannot exceed MusicBrainz or other declared external-service limits; selected subtitle downloads also run in a foreground Entity lane instead of holding an API request open.
- Newly imported videos now prioritize technical probing and subtitle reconciliation so they become playback-ready ahead of unrelated background work, and preview/trickplay jobs wait for probe metadata instead of surfacing false scan failures.
- Wide 4K video masters such as 3840×1920 are no longer mislabeled as 1440p in library thumbnails and web metadata when playback correctly identifies them as 4K.
- The launch-site navigation drawer now renders reliably above page content on
  mobile Safari.
- Launch-site section links no longer all appear selected at once in desktop and
  mobile navigation.
- Plugin metadata can no longer stamp one parent or locator identity onto several structural children: shared proposal identities are discarded before matching or persistence, and upgrading repairs affected AniList episodes, MangaDex volumes and chapters, Open Library series children, and repeated MusicBrainz track recordings already stored in the library.
- Monitored container sync now keeps exact Entity bindings through nested provider hydration, preventing contextual parent aliases from aborting a season/episode merge after the Request flow has already matched or created the wanted children.
- Applying new provider metadata to a monitored container now immediately checks that exact Entity for provider-only missing works, so a season with no files on disk becomes Wanted without waiting for the daily sweep. Targeted monitor checks also run independently from the global sweep and other Entity checks.
- Video pages now show the saved resume position on the seek bar immediately, before the stream finishes loading.
- On-demand monitored checks now queue immediately for only the selected Entity, continue updating its detail page while provider work runs, and cannot be blocked by a different monitored item's metadata failure. Season activity also rolls up searches and downloads from episodes underneath it, making continuously numbered provider seasons visible while they work.
- Downloads now wait without blocklisting when their client is offline or unhealthy, recover cleanly when a client is removed from settings, and return to a fresh Wanted search when the client-side download is deleted.
- Turning Monitor on now starts an immediate provider sweep with editable quality-profile and import-target choices, promotes provider-known missing works into visible Wanted items, and searches gaps throughout owned container trees. Provider matches also reuse oddly titled local season folders instead of creating a duplicate canonical season.
- Entity acquisition history now retains import failures recorded during manual-review holds and import exceptions, so those events no longer disappear from the affected item’s history.
- Video scans now keep unnumbered, custom-titled season folders inside an established parent series and order them by title after detected numbered seasons, instead of promoting each folder to a separate series.
- Video remuxing now preserves a negotiated multichannel audio track instead of unnecessarily downmixing it to stereo AAC.
- Downloads and entity detail pages now retire passive requests once their surviving entity is fully present, while preserving partial imports and deliberate upgrades; refreshes are serialized and source checks stay scoped to visible entities so large libraries no longer saturate PostgreSQL.
- Long titles, release paths, and messages in Request History now wrap within their rows instead of overflowing the page.
- Reconciled imports now retire older requests and monitors for the same provider identity, passive requests whose target was removed no longer render as broken Downloads cards, and acquisition cards restack before their status and action controls can overlap.
- The Downloads view now refreshes attention-only requests and rechecks when the browser becomes visible, so a scan that reconciles an awaiting-selection item into the library removes its stale request card without a manual reload.
- Partial music imports now keep provider tracks visible as missing and non-playable, expose per-track monitoring, immediately search for the remaining tracks, merge equivalent Soulseek filenames without duplicate library entries, and survive concurrent library scans without losing newly imported files.
- A slow indexer's HTTP timeout now stays scoped to that indexer instead of cancelling and requeuing the entire acquisition search; usable results from Soulseek or another source proceed immediately without repeating broader queries against the failed source.
- Soulseek music matching now understands artist/album/track path-shaped results and treats quotes, slashes, and other punctuation as token boundaries, so valid releases such as quoted soundtrack singles are no longer mislabeled as title mismatches.
- Wanted audio-track placeholders remain visible in acquisition monitoring but are excluded from playable album, track, collection, restored-queue, and native browse surfaces; the audio source API also refuses to stream them until a real file fulfills the request.
- Completed acquisitions now preempt remaining release searches, so a large artist request cannot leave finished Soulseek downloads waiting to import while its metadata and search fanout continues.
- Manual replacement reviews can now be retried safely after a failed download-client handoff: Prismedia reuses the reviewed upgrade and candidates, repairs old transferless Preparing rows into an actionable state, and prevents same-request duplicate replacement children.
- Imported singles now attach their downloaded audio file to the original metadata track—even when filenames include track numbers or bitrate decorations—preserving provider identity and Wanted fulfillment. Existing scanner-created duplicates heal on the next audio scan without losing probe data or fingerprints, and entity-bound waveforms regenerate against the retained track. Their probes run at import priority, while foreground search floods remain confined to reserved worker capacity so standard jobs keep advancing.
- Legacy slskd enqueue errors now reconcile the exact peer files before Prismedia decides whether the handoff succeeded, failed cleanly, or remains ambiguous, preventing a rejected peer from leaving requests permanently stuck in Preparing or Stopping.
- Failed Soulseek transfers now enter blocklist-and-retry recovery immediately instead of waiting behind a large artist's remaining searches.
- Acquisition transfer polling now runs ahead of large request search fan-outs, so completed or failed Soulseek downloads keep advancing through import and recovery while the remaining artist albums and tracks are still being searched.
- Soulseek album searches can now persist opaque multi-file release locators of any practical length instead of failing the entire acquisition when a peer folder encodes beyond the former URL limit.
- Artist requests now record monitoring for every selected album before background metadata hydration begins, so the child-activity panel keeps all requested albums visibly monitored and labels the pre-search phase as Preparing metadata instead of making queued work look turned off.
- Artist requests now retain the reviewed album artwork URLs in the initial library graph, so album cards reuse the thumbnails already shown during review while background hydration localizes the image files.
- Soulseek searches now serialize through slskd's one-operation limit and wait for each asynchronous search to finish before reading peer responses, preventing HTTP 429 backoff and valid album or track results from being discarded in progress.
- Soulseek downloads now fall back to slskd's legacy per-user queue API when the configured server predates batch downloads, while retaining grouped progress and import paths for albums and individual tracks.
- Individual audio-track acquisitions now pass through the same crash-safe library placement checkpoint as album imports and bind the downloaded file to the existing wanted track, album, and artist graph instead of failing during or after placement.
- Fixed large artist requests waiting on every album acquisition or cover download before returning: reviewed album entities, metadata, and provider routes are committed in one batch, while a durable background fan-out expands each album's track graph before publishing its acquisition search.
- Album searches that find no acceptable complete release now start artist/album/track-qualified child searches immediately. Track hydration happens before the parent search, existing barren track acquisitions are requeued instead of counted as finished, and repeated album/track titles collapse to one search term, so individual-file sources such as Soulseek no longer wait for a later enrichment or monitoring pass or receive phrases like `Artist Title Title`.
- TV imports now reconcile byte-identical episodes that already reached the library—whether or not a concurrent scan indexed them first—instead of creating duplicates, demanding manual import, or blocklisting a release that already satisfied the request. Imports waiting behind a library scan also show that state explicitly.
- Monitored series, authors, and artists now bypass cached provider proposals whenever checking for new releases and preserve each discovered work's plugin-declared identity across hierarchy levels (such as TMDB series → season → episode), so newly published seasons and episodes appear on the next tracking pass instead of being discarded.
- Collection grids no longer crash when separate members resolve to the same artwork, and member accounts no longer wait tens of seconds for library visibility to be recalculated throughout every browse query.
- Collection grids no longer stall while resolving fallback member artwork in large libraries; visibility checks are now scoped to the collection's actual members.
- Request searches and metadata enrichment now use the same foreground priority as manual Identify work, while completed downloads enter the import pipeline ahead of background scans.
- Movie and TV imports now catalog only the files they placed instead of scheduling a full video-library scan after every completed download, preventing season-pack imports from clogging the queue.
- Imported TV episodes now run Auto Identify and subtitle processing against only the episode that was placed, avoiding a series-wide metadata cascade after every download.
- Stream-copy video playback now preserves delayed audio-track timestamps without re-encoding already-compatible AAC audio, keeping the source's original surround samples, bitrate, and channel layout intact while preventing lip-sync errors in web and native clients.
- Playback Stats now show only the signed-in user's activity by default, so legacy household history and other members' plays no longer appear in personal totals, charts, top items, or recent events. Administrators can switch the page between any one account and an explicit All users view; members remain locked to their own activity.
- Wanted entities now follow the library access of their request target, including legacy placeholders that fall back to the media kind's default profile, so household members cannot browse or open requests destined for libraries they cannot access.
- Identify now matches child titles through filename-style separators, punctuation, apostrophes, and accents, so multi-disc albums and similarly structured seasons can map their local children even when per-disc or per-season positions repeat.
- tvOS compatibility players can now request malformed HDR sources for local tone mapping and direct playback instead of forcing a server video transcode.
- HDR video with contradictory sub-10-bit samples is now normalized through the SDR tone-mapping path instead of being copied with HDR signaling that makes standards-compliant TVs render it severely dim.
- Acquisition imports now remain in the durable queue and enter the CPU/disk-intensive import pipeline one at a time, rather than claiming many completed episodes as simultaneously running while they wait for the same media-placement coordinator.
- Season Acquisition tabs now open active episode work automatically and show each episode’s search/download state independently from its future-monitoring switch.
- Download-client handoffs now recover automatically after a lost response or interrupted add instead of leaving acquisitions stuck on “Preparing download”; qBittorrent adds that conclusively create no identifiable torrent now release the failed attempt and immediately schedule monitored child fallback instead of waiting for the next daily retry.
- Season acquisition pages now keep episode activity mounted during live refreshes, show searching/downloading/importing counts directly, and demote a failed season-pack attempt while episode fallback is active instead of flashing or presenting the stale parent failure as the current outcome.
- Prowlarr aggregate release searches are limited independently from general worker concurrency, preventing large episode-fallback batches from timing out while leaving scan, probe, import, and other worker throughput unchanged.
- Monitored season requests now fall back to exact per-episode searches when no acceptable season pack exists, every attempted pack download fails, or a downloaded pack cannot be imported; selectable and in-flight season packs still finish without duplicate episode work.
- Acquisition imports now re-identify the source-backed entity after its real files and children are cataloged, so organized Wanted placeholders keep their stable provider identity while gaining episode titles, track names, and other child metadata.
- Interrupted acquisition imports can now be explicitly started over: Prismedia removes their partial files and stale catalog state, restores recoverable pre-upgrade files, and creates a clean search. Removing a download uses the same graceful cleanup even when its download client or payload is already missing.
- Flyouts, dropdowns, popovers, and dialog panels now use opaque neutral surfaces so underlying content no longer shows through their controls.
- Audiobook playback now uses its compact progress timeline without generating an oversized, low-information waveform, and unusually long audio waveforms are bounded safely instead of overflowing the browser canvas.
- Restored EntityGrid toolbar floating behavior so its search and controls remain fully visible beneath the sticky app header while scrolling.
- Audiobook parts now stay scoped to their Books instead of appearing as music in track lists, global activity, smart or manual audio collections, Jellyfin music catalogs, and playback statistics.
- Full book-library reconciliation now preserves audiobook chapters on the Book explicitly targeted by an acquisition, even when its readable edition lives elsewhere, instead of splitting the audio into a duplicate Book after import.
- Audiobook listening progress now uses browser-learned chapter durations while audio probing is still pending, so multipart resume and completion save correctly from the first play.
- Books now manage ebook and audiobook renditions independently: requesting the missing counterpart reuses the same Book, the Acquisition tab shows both states, and stopping one rendition leaves the other untouched. Open Library requests also no longer fail with a false "proposal changed" conflict.
- Native clients now correctly hide NSFW libraries and file directories when their content-visibility setting is disabled.
- Download profile settings now respect the viewer's library access and current SFW/NSFW mode; profiles targeting an NSFW library are hidden whenever NSFW content is hidden.
- Library scans no longer remove wanted or actively monitored movies, seasons, and series while they are waiting for files.
- Adding or updating a library with a folder that is already watched now returns a clear conflict instead of failing with an internal server error.
- Video seek bars now span the same full player width as the playback controls, regardless of the media artwork or source aspect ratio.
- File deletion now lives in the shared Acquisition tab for every file-backed Entity instead of crowding the detail hero, whose remaining actions wrap cleanly on smaller screens.
- The video film strip now responds only to horizontal wheel or trackpad movement, so ordinary vertical page scrolling passes through instead of scrubbing playback.
- Identify now operates only on Entities with real source media on disk: Wanted and fileless metadata placeholders no longer show Identify actions, cannot enter manual or automatic queues, and stale queue work is retired without duplicating matching Wanted children during a parent cascade.
- Deleting the wrong files from a monitored item now replaces its completed acquisition with a clean retry and returns it directly to Searching, instead of leaving an Imported card backed by a deleted acquisition that failed with "Acquisition was not found."
- Managed Entity deletion now requires deleting the source files too, preventing library-only removals that a scan would immediately rediscover; committed deletions also remain successful when immediate monitored recovery must wait for the scheduled sweep.
- Remove wanted now uses the same durable monitor and download cleanup as Unmonitor; items stay visible and selected when remote cleanup fails or an import race retains them, so the UI never claims they were removed when they were not.
- Book, movie, TV, and album imports now bind every placed file to its Entity graph before reporting Imported, so newly acquired content is readable or playable immediately instead of waiting for a later library scan; background scans remain reconciliation and cannot turn an already-ready import into a false failure when housekeeping is unavailable.
- Book, movie, TV, and album imports now checkpoint their exact file placements before touching the download, then resume those same targets after a crash or interruption without losing move-mode payloads or creating duplicate copy/hardlink suffixes; cover art and other carried companions recover with their media.
- Library scanning and import now follow the host filesystem's path-case rules, so case-distinct files and folders remain separate on Linux while Windows keeps its case-insensitive behavior.
- Updated the OpenAPI runtime to a patched release that rejects the known circular-schema denial-of-service vulnerability.
- The shared entity API now honors the Wanted filter, and the generated web client includes the current acquisition volume field and wrong-volume rejection code.
- Entity descriptions now render through a closed Markdown renderer that blocks raw HTML and unsafe link schemes, preventing plugin or edited metadata from injecting executable content into detail pages.
- Fixed the whole app intermittently failing with server errors (500) under background-job bursts: the bundled PostgreSQL allowed only 40 connections while the API and worker connection pools could together demand far more. The pools are now capped conservatively and the bundled database accepts 120 connections (existing installations are upgraded automatically on restart).
- Fixed live download telemetry (speed, seeders, peers, size) never appearing anywhere in the app: three qBittorrent per-torrent calls (properties, file list, piece map) used the wrong query parameter and were silently rejected. This also un-breaks seed-goal tracking and the mid-download wrong-content check for torrents.
- Fixed detail pages visibly flashing back to their loading skeleton when an acquisition action (search, cancel, delete-revert) refreshed them — they now update in place.
- Deleting library items now also removes their generated assets (previews, trickplay, waveforms, and — on full removal — grid thumbnails and downloaded artwork) immediately instead of leaving orphaned cache files behind.
- TV imports now align "as aired" multi-episode files by their episode TITLES: a pack whose file labeled E01 actually carries provider episodes 1 and 2 (and E02 carries 3 and 4) is placed at the right provider numbers, and the covered second episodes are bound to the same file so they play instead of staying wanted forever. Conflicting or unmatched titles fall back to the file's own numbering — never a guess.
- Fixed Jellyfin-compatible clients taking up to a minute to load the initial library/collections view: box-set rows previously hydrated every member of every collection just to decide playlist tagging; they now use one grouped query for the whole page (and gain proper child counts), and paged library browsing (Infuse's scrolling) now fetches only the requested page window instead of materializing the entire library on every call. Single-item detail remains fully hydrated.
- Fixed album and artist grids feeling slow: audio library scans now backfill the small grid-thumbnail variants for album/artist covers (and re-scans self-heal missing ones) exactly like video scans do, instead of serving full-size original artwork until the daily sweep reached them.
- Comic and volume-scoped book searches now guard against wrong volumes: a Volume 1 request rejects releases naming a different volume ("Vol 3", "v03"), prefers releases naming the exact volume over unlabeled ones, and volume-less releases (omnibus editions) stay available. The same protection wrong seasons already had for TV.
- Automatic release selection for music and books now verifies work identity like movies and TV: a music search only auto-grabs releases naming the exact artist and album (remaster/deluxe editions still pass — years are deliberately not a rejection axis for music), and a book search requires both the exact title and the author in the release name, so a same-title book by a different author can no longer win on quality or seeders. Rejected releases stay visible for manual picks.
- Fixed single-episode searches rejecting multi-episode releases that contain the sought episode: a double-episode file like S01E41E42 now fulfils a search for either half (and dashed ranges like S01E01-E03 count for every episode they span).
- Fixed automatic release selection grabbing the wrong same-name work: a movie or series search now verifies the release actually names the sought title and rejects releases whose title-adjacent year points at a different remake or reboot (the 2019 "Clifford the Big Red Dog" reboot no longer wins a search for the 2000 original just because it exists in higher quality), and releases naming a longer different title (sequels, spin-offs like "Dune: Part Two" against "Dune") are rejected instead of merely outranked. TV searches also now compare against the series' première year from your library rather than unreliable request-time metadata.
- Fixed newly imported videos sometimes failing playback with an HLS playlist 404 before their background probe metadata was saved.
- Fixed one problematic file freezing an entire library scan: a file that fails to save is now skipped and reported by name while every other file persists normally, the failing file alone is retried on the next scan (instead of the whole library rescanning or new imports never appearing), and in all-libraries scans one broken library no longer stops the remaining libraries from scanning.
- Fixed the Jobs dashboard mislabeling acquisition, backup, thumbnail-sweep, and other newer job types as "Relocate video generated files / Library maintenance": every job type now has a proper name, description, and queue grouping (acquisition work gets its own Acquisition group), per-queue cancel and clear-failures now reach those jobs, and the SFW-mode privacy masking of real maintenance jobs uses an honest generic label instead of inventing a fake file-moving task.
- Fixed a corrupt or truncated media file grinding the job queue forever: probing it now records the file as unreadable and fails once with a clear message naming the file, instead of silently succeeding while preview generation failed and re-queued itself on every scan. Scans stop scheduling repeated probe/preview/trickplay work for a file marked unreadable while still allowing adjacent subtitle files to reconcile, and repairing or replacing the video automatically gives it a fresh chance.
- Fixed the hourly collection refresh failing on every run (and dynamic collections going stale) when a collection rule filtered on play count or skip count: the rule engine still queried the pre-multi-user playback table that the accounts upgrade removed. Those rules now read the per-user watch state and match when any household member's count qualifies.
- Fixed release matching for indexer titles that use underscores, dots, or hyphens instead of spaces: acquisition scoring now recognizes those separators for quality, format, language/proper tokens, profile terms, title-only blocklists, and closer title matches so spin-offs or subtitles are less likely to beat the exact requested work.
- Fixed detail pages showing a different cover than the library grids for the same item: when an item had both downloaded artwork and a cover extracted from its file (common for identified books), the detail poster skipped the downloaded artwork the grids show and fell back to the embedded one. All surfaces now agree on the same winning cover.
- Fixed MangaDex request results showing the MangaDex "read at" placeholder instead of real cover art by applying the same no-referrer image loading behavior used by Identify to shared request thumbnails and detail artwork.
- Fixed Safari showing placeholder icons instead of cover artwork across library grids and shelves: WebKit fires a spurious image error for lazy-loaded covers using the `sizes="auto"` responsive hint, so cards no longer emit it and Safari renders the same small grid variants as every other browser.
- Fixed library scans crashing — and never completing again — once a library contained a multi-episode file (e.g. `S01E05-E06`) bound to both episodes it covers. Every scan failed with a duplicate-key error, which also blocked freshly imported episodes from ever appearing in the library until the offending scan succeeded.
- Fixed Jellyfin clients (Infuse and friends) taking 30+ seconds to open a series library: list rows no longer load each series' full metadata graph one by one — a 50-series page went from ~9,800 database queries to a handful of batched ones, and the Continue Watching / Latest shelves shed a redundant double-load per item. Opening a single item is unchanged and stays fully detailed.
- Fixed a half-downloaded season page not offering the missing-episode search: seasons now roll up their wanted episodes with the same "Search N missing" action the series page has.
- Fixed the Jobs dashboard costing seconds per refresh on long-lived installs: queue counts are now aggregated in the database instead of loading the entire job-run history (hundreds of thousands of rows) into memory on every poll.
- Fixed the Files page hanging up to a minute when opening details on a huge folder: the file-count/size walk is now time-boxed, and a tree too large to measure quickly simply omits those stats instead of stalling the page.
- Fixed a monitored series going quiet with episodes still missing: when a season pack imports incomplete, Prismedia now searches for each missing episode individually (Sonarr-style) — every gap becomes its own monitored, auto-grabbing episode acquisition — instead of marking the season done with holes.
- Fixed "Search for release" doing nothing on a wanted series page: requesting a series now requests each of its still-wanted seasons, skipping ones already in flight.
- Fixed entity detail pages taking seconds to open: identify, monitor-eligibility, and request lookups no longer fetch the community plugin index from the internet on every page view (that index is only consulted by the plugin browser, and is now cached), provider metadata lookups are cached for a short window so re-opening the same series is instant, and a series page paints immediately instead of waiting for the provider's season list.
- Fixed expanded acquisition history rows overflowing phones when release titles are long; the history now wraps inside the acquisition card instead of widening the page.
- Fixed the acquisition settings and release picker not adapting to phones: indexer, download client, path mapping, custom format, profile, and blocklist rows now stack as readable cards on narrow screens instead of clipping off the edge, and the release picker shows its stacked cards whenever it actually lacks the room for the full sortable table — including on tablets and in narrow panels — rather than deciding by screen size alone.
- Fixed one marker-less release winning multiple TV season searches: season and episode searches now rank unit precision above quality — a release naming the exact season (or episode) outranks a complete-series pack, which outranks a title with no season token at all — so the same generic DVD rip can no longer be auto-grabbed for four different seasons.
- Fixed the misleading "torrent did not appear in the category" failure when qBittorrent silently ignored a duplicate add: the client now recognizes the already-present torrent by name and reuses it, or reports an honest duplicate error; the connection test also verifies the download category can actually be created, so real category problems surface at setup time.
- Fixed automatic downloads giving up when the best release could not be queued: auto-pick now tries the next-best accepted releases before falling back to manual selection, and never leaves an acquisition failed while pickable releases remain.
- Fixed NSFW libraries being granted to accounts that block NSFW content: the upgrade migration now skips NSFW libraries for previously-SFW profiles, every grant path enforces the same wall server-side (an NSFW-blocked user can never hold an NSFW library grant, and blocking a user's NSFW permission revokes any they had), and the access editors hide NSFW libraries for blocked accounts.
- Fixed monitoring an existing on-disk series (or movie/album) duplicating it on the next import: an acquisition linked to an entity that already lives on disk now merges into its real folders — new episodes land in the existing season folders (a missing season gets a template-named folder inside the existing series folder), a strictly-better release replaces the owned file in place (with the previous file kept recoverable), and files that aren't upgrades are dropped. A release with nothing better than what you own fails with the release blocklisted so it is never grabbed again; an upgrade that would change the file format is held for manual import instead. Your existing folder layout is never renamed or reorganized.
- Fixed acquisition-imported shows and albums arriving with no metadata: the import now stamps the provider identity (e.g. the TMDB id) onto the imported entities and kicks off identification for the affected series/artist/movie right after the scan — even when Auto Identify is switched off for the library, since a requested download is explicit intent.
- Fixed a fresh install landing on the sign-in page instead of the setup wizard when the first status check failed: setup detection now retries through boot-time hiccups, never guesses "setup done" from a failure, and the sign-in page itself forwards to the wizard whenever setup is still pending.
- Fixed book imports failing when no book acquisition profile exists: the import now uses the promised permissive defaults (the first book-enabled library, the default naming layout, a move import) exactly like movies, TV, and music already did.
- Fixed Availability filtering counting request-time cover art as source media: a Wanted Entity can carry a downloaded cover while still correctly remaining fileless, and **On disk** now means verified source ownership only.
- Fixed release searches that could sit "searching" forever after an interruption (e.g. a server restart mid-search): a stuck search is now detected within about a minute and marked failed with a clear retry message, and indexer requests carry explicit timeouts.
- Fixed removing a download from the Downloads tab erasing the item's Wanted state: removing a (typically failed) download now only cleans up the transfer — the wanted item stays in your library, and cancelling from the item's own page remains the way to give up on it.
- Fixed the Request page's tab bar overflowing the screen on phones: the tabs now scroll within their own row instead of widening the whole page.
- Fixed the Books/eBooks library grid so books grouped under an author still appear in the main Books view while true nested book children stay on their parent detail page.
- Fixed many playback reliability issues, including first-play failures, slow or inaccurate seeking, stalled transcodes, remux segment errors, HDR/Dolby Vision handling, subtitle rendering, audio-track selection, progress bars, cache cleanup, and high CPU spikes.
- Fixed progress and history consistency so Prismedia and Jellyfin-compatible clients agree on resume points, completion, play counts, skips, watched/read state, and Start Over behavior.
- Fixed Identify so searches, provider retries, child cascades, metadata merges, artwork selection, NSFW propagation, duplicate provider data, music matching, and bulk accept behave reliably without losing existing metadata.
- Fixed scan and worker resilience, including startup ordering, stale running jobs, cancellation races, recurring scan timing, deleted/disabled libraries, missing cache assets, failed-job recovery, and Docker build stability.
- Fixed Jellyfin-compatible client setup, login, browsing, shelves, similar-item probes, collection roots, direct-child series, episode artwork, music sync, cast navigation, and playback metadata.
- Fixed OPDS edge cases for long client identifiers, reverse-proxied HTTPS links, folder-backed comics, series containers, covers, and acquisition feeds.
- Fixed mobile and touch interactions across navigation, library menus, thumbnail scrubbing, lightboxes, readers, detail heroes, forms, buttons, flyouts, and narrow-screen wrapping.
- Fixed image, gallery, animated-image, feed, and lightbox behavior so inline clips, GIF/APNG files, portrait media, previews, preload, and aspect ratios render correctly.
- Fixed sub-view grids so albums, galleries, series sections, linked content, and collection previews expose the same selection and Add to Collection controls as main library grids.
- Fixed collection grid cards so they inherit member artwork on list pages, matching detail-page covers.
- Fixed dynamic collection rules, previews, saved edits, smart collection queries, random sorting, page-level grids, tag/person/studio references, and orphan cleanup controls.

### Removed
- Removed the obsolete frontend scraper/package-manager, execution-result, and StashBox panels; the Plugins page now uses only the current generated provider catalog, install/update/remove, Stash community, and credential contracts.
- Removed the unused TypeScript plugin SDK packages and the TypeScript/Python plugin authoring guides; plugins run on the .NET runtime, and Stash YAML scrapers remain fully supported through the built-in Stash-compat engine.
- Removed the app-level API key and the API Access settings panel. All access — web, Jellyfin-compatible, OPDS, and future native apps — is per-user with usernames, passwords, and session tokens. On upgrade the old key lives on only as the migrated accounts' initial password.
- Removed LAN auto-login: browsers on the private network no longer receive an authentication cookie automatically, and the LAN visibility auto-enable setting is gone. Sign in once per browser instead.
- Removed Stash-compatible perceptual hashing (pHash). Fingerprint workflows now rely on MD5 and OpenSubtitles hash support, and the old pHash setting and StashBox pHash documentation are gone.
- Removed development-only route shells and redundant prerelease UI panels so the app surface and release notes focus on the production Prismedia experience.

### Docs
- Prepared the native App Store listing, exact-size iPhone, iPad, and Apple TV
  screenshot galleries, review notes, privacy answers, licensed
  demonstration-media ledger, and an isolated Apple review deployment that
  cannot mount the household library.
- Added public support and privacy pages for the native app listing, including
  self-hosted data boundaries, reviewer-server handling, and direct support paths.
- Reworked the repository README around Prismedia's launch story, product film,
  native app experiences, Product Hunt page, and campaign artwork.
- Added wide, square, and vertical Product Hunt campaign artwork, strengthened the
  maker profile around Prismedia’s privacy-first native product story, and added
  reproducible checks for social-asset dimensions, copy clipping, and upload size.
- Scheduled Prismedia’s Product Hunt launch for August 23, added the official launch badge to the public site, and documented a preparation, launch-day, and follow-up campaign.
- Added a reproducible Product Hunt launch kit with final listing copy, a square prism thumbnail, and six visually reviewed 1270×760 gallery cards covering the media lifecycle, native playback, reader customization, purpose-built media experiences, and self-hosted architecture.
- Prevented the web-platform proof frame from creating horizontal overflow on narrow launch-site viewports.
- Reframed the public README around the connected media lifecycle, product film, TestFlight path, native reader customization, and full-fidelity Apple TV playback.
- Added a dedicated social sharing card, richer product discovery metadata, and an explicit crawler entry point for the public launch site.
- Refined the launch site with full-spectrum headline accents, clearer native-platform composition, an explicit film play control, transparent optical marks, and full-fidelity native playback messaging.
- Extended the launch film's one-light-in, spectrum-out story across the product site with a labeled media map, centered optical paths, media-specific experience rails, platform accents, a self-hosting topology, and a responsive closing call to action.
- Rebuilt the documentation landing page as a product-first launch site with real web, iPhone, and Apple TV captures, an embedded 72-second silent product film, explicit video, reading, audio, and image experience stories, and direct install and TestFlight paths while preserving task-oriented Docs routes.
- Documented audiobook scanning and manual placement, including supported formats, Books-root configuration, folder grouping, ebook matching, multipart ordering, and rescan behavior.
- Replaced the README, documentation-site theme, brand assets, and product screenshots with the restrained prism design language across current desktop and mobile surfaces.
- Overhauled the README and documentation site around the current Prismedia app, including setup, browsing, scanning, playback, requests, identify, collections, settings, operations, plugins, and screenshots.
- Added Library & Scanning references with supported folder layouts and file extensions for videos, movies, series, images, galleries, comics, ebooks, and audio.
- Added Jellyfin Compatibility docs covering tested clients, sign-in profiles, API keys, SFW/NSFW visibility, setup flows, and reverse-proxy bypass rules for Authelia and Authentik.
- Added deployment and security docs for Docker, authentication, API keys, reverse proxies, release-channel images, and the .NET API serving the built Svelte app.
- Added developer-facing architecture and flow docs with diagrams for app boundaries, generated clients, jobs, identify, playback, and release-readiness review.