tristenlammi/Arrmada v0.4.0 on GitHub

Highlights

Shows with S01-style season folders, or kept inside a category folder, work fully
again (#56). New episodes go into the folders the show already has, and files v0.3.0
and v0.3.1 put in the wrong place are listed on the show's page to put back.
"Released" now waits for the digital or disc release, not the cinema date, when
almost every release is a cam. Films show their cinema, digital and physical release
dates, and the Calendar lists the digital and disc releases (#93).
Untagged audio tracks are identified (#90). Where the local AI (whisper) is set up,
Arrmada listens to tracks tagged und or nothing, so a Must audio language can judge them.
A global seed goal (#78) for indexers set to use it, never applied to anything
you already have.
The disk guard works per drive (#70). With two download clients on different drives,
only the full drive's downloads pause. It can also measure a different drive, for Unraid
cache pools and ZFS.
Manual import suggests the downloads that match, and can browse or search any
downloads or library folder you gave Arrmada.
Imports wait for qBittorrent's move (#94). A season pack is no longer blocklisted and
grabbed again while qBittorrent moves it out of its incomplete folder.
Two shows with the same name no longer get each other's episodes (#81), and episodes
named like 1x8 are read.
A show's search fills from releases you don't avoid first, so a starved
complete-series pack no longer beats good season packs.
Arrmada never throws away a video it doesn't track. Convert and imports keep it, and
a film's page shows +1 other file when its folder holds a video Arrmada doesn't use.
Music (Preview) is rebuilt: editions, a matcher that scores downloads like beets,
single-file albums split by their cue sheet, clean tags and naming, new releases followed,
and a fast library scan.
Books: KEPUB downloads for Kobo, Calibre libraries scanned from metadata.opf, and a
book's details editable by hand.
SABnzbd and Prowlarr on Unraid, built into the Unraid container and off until you
switch them on. There's also a guide to downloading through a VPN with Gluetun (#92).
A round of security fixes from a full audit, covering sign-in, saved passwords and keys,
the bundled download clients and file handling, and Arrmada is built with Go 1.26.10.
Some of the fixes change behaviour, so read Before you update.

Before you update

"Released" now waits for the digital or disc release (the cinema date + 90 days when
neither is known). It's the default for new films and requests, and every film already
on Released waits too: the scheduled search, RSS, search on add and a request approval's
search wait for that day. Search on a film doesn't. To keep grabbing at cinema time,
set the film's minimum availability to In cinemas. See Changed below.
A release that doesn't say which of two same-named shows it's for is no longer grabbed
automatically (#81). This only matters for a show that shares its title with another in
your library (Scrubs 2001 and 2026). A year-less release that could be either is skipped
by searches, and a grabbed one waits in Needs you as Which show? instead of
importing. See Fixed below. Separately, a library scan no longer guesses which show a
year-less folder is when more than one show has its exact title: it goes to review.
A rescan no longer replaces an episode's file with one that isn't named as the show,
and the duplicate check no longer offers the show's own copy for deletion beside it.
Episodes an earlier rescan already switched to another file aren't changed by the
update: check them and import the right file by hand. See Fixed below.
Untagged audio tracks are listened to in the background (#90). It's on by default
wherever a whisper model is installed, and uses the GPU or CPU whisper uses, one file at a
time and never alongside a subtitle. Turn it off under Settings → Subtitles → Local AI
(whisper) → Identify untagged audio. Without whisper it does nothing.
Imports wait for qBittorrent to finish moving a torrent (#94). With qBittorrent's
incomplete-torrents folder on, a finished download imports once it's in its final folder.
A download folder Arrmada can't read is held and shown in Needs you, never
blocklisted or deleted.
An import never replaces a file no record owns. It lands beside it with the version's
label (Heat (1995) - 1080p BluRay - Default.mkv), and a conversion whose new name is
taken waits under Convert's Problems until you rename or remove that file. See Fixed below.
Refresh & rescan keeps a film's file while it's on disk. It used to switch a film to
the biggest video in its folder whenever that was a different file, even an extra
version's file, a trailer or another film's, so two records could hold one file and
removing the version recycled the film's file. Now:
- While the recorded file is on disk (or can't be checked, as when a network share is
  down), Refresh leaves the film's file as it is. A bigger copy beside it shows as
  +1 other file on the film's page instead.
- To switch a film to another file in its folder, remove or move the old file, then
  Refresh: the biggest video left in the folder (or two folders down) becomes the film's
  file. This works only in a folder that is plainly the film's own (named for it, with no
  other film's file in it); anywhere else, import the new file by hand.
- It never picks a file another version or film records (a hardlink of one included), a
  sample, a trailer, featurette or anything in an extras folder, anything in Arrmada's
  hidden folders, or a file in a subfolder named as another film.
- A film with several versions whose main file is gone has that record cleared, not
  replaced by one of its versions' files. Its other versions keep their files.
- A film loose in a Movies folder, or in a folder it shares with other films, whose file
  is gone has the record cleared; nothing else in that folder is taken for it.
- If the folder can't be read, or holds too many files to look through, nothing changes.
- A file renamed only in case (Heat (1995).mkv to heat (1995).mkv) is still found.
- A recorded file under 50 MB is no longer cleared as if there were no video.
What a show's rescan and duplicate check read is narrower (#56). They no longer read a
folder inside the show's that holds another show (hidden, nested, or under Other or
Bonus), Plex extras folders, a file named as another show in your library, a video
loose in the folder of a show kept in season folders, or a file with no title in its name
outside the folders the show already has files in. A show with no files yet reads only
its own Title (Year) folder. If a rescan used to attach one of these, attach it with a
hand import instead.
A show whose folder also holds another show's episode waits in Needs you (#56). An
episode of another show loose in the show's folder (Peppa Pig - S01E01.mkv in
Kids), another show's folder, or a link inside it means the show's folder can't be
told: downloads wait in Needs you, and Refresh & rescan, Rename, the duplicate
check, Move to folder and Map files… are refused for it (Map files uses the same
check as the import). Move the other show out, then use Refresh & rescan.
Move to folder refuses a show folder holding an episode the show doesn't have (#56).
If anywhere in the show's folder there's an episode Arrmada hasn't recorded for it whose
name doesn't say it's this show's (Other Show - S01E03.mkv in its S01 or Season 1),
or one naming another show in your library, the move is refused and the message names the
file: moving the folder would carry it along. Import it, or move it out, then move the
show. Themes, backdrops and trailers don't count; a small file or a sample named as an
episode does.
The way out for a show whose files were moved now needs a folder named with the
show's year (Show (2019)) or one already holding its files; see Fixed below.
A Must audio language now has to be in the release name (#58). A release is grabbed for
a Must language only when its name says it (ITA, iTALiAN, VFF, TRUEFRENCH…), or when the
language is the title's original and the release names no language or is MULTi/DL/Dual
Audio, which always carry the original. A MULTi or Dual Audio release that doesn't name the
language used to count as "might have it" and was grabbed; it's now skipped, and interactive
search says why ("Doesn't name Italian audio…"). If your trackers label releases MULTi when
they carry your language (French trackers often do), turn on MULTi and Dual Audio
releases count as having a required language under the profile's Audio language row.
It's off on every existing profile. A release that names no language, for a title whose
original language isn't known, is still not refused. This changes which releases are
grabbed only: files already in your library are judged by the same rules as before (the
newly read spellings below count for releases, not for library files).
Hong Kong films get an original language. TMDB codes Cantonese as cn (or yue),
which Arrmada didn't know, so those films had none and the Audio language rules skipped
them. They now read as Chinese: "Original language" rules apply to them, an untagged
release of one counts as Chinese (so a Must English refuses it, as for any foreign film),
a profile that requires the original marks such a film whose file has only a dub as not
fitting, and one that prefers it can upgrade that file, within the per-sweep upgrade
limit. A file whose audio is tagged or heard as Cantonese now reads as Chinese too, so an
English Must or Prefer can upgrade it (within the same limit). TMDB's xx (no spoken
language) is treated as no original language.
Missing music files are flagged, not downloaded again. When a track file is gone on two
checks a day apart, the album says "N track files gone from disk" and isn't searched by
itself, like movies. Use Search for missing tracks on the album page, or Music → Library
tools for all of them at once (each run takes 25 albums, the ones not searched yet first, and
asks first). Moved or retagged the files yourself? The check then starts a library scan, which
links them again at their new place instead of downloading. A folder that looks empty or
unmounted is skipped and named in a health note; if the music folder itself looks unmounted, or
too many files vanish at once, nothing is marked at all. An album on a share that's away is never
re-linked to another copy of it.
A music library scan of an empty or unmounted music folder is refused and changes nothing.
Untagged folders matched only by their name are placed only as a whole album (or exactly the
tracks still wanted); anything else goes to Reviews.
Music upgrades are off until you turn them on. The update switches automatic upgrades off
on every music profile, and new profiles and presets start with them off. Turn them on per
profile (Quality → Music → Automatically upgrade).
Artists with no music profile now use "Lossless, or the best MP3", created once if you have
no music profile. They were on "Any quality" before, so MP3-192 and below, OGG and Opus are no
longer grabbed for them. Pick another profile on the artist to change it.
Adding an artist searches its wanted albums straight away (up to 25, in the background),
following Settings' "search on add". The add dialog has a "Start searching now" box.
Homepage widget: wanted.total now includes albums (released albums still missing) while
Music is on. New fields artists, albums, tracks and wanted.albums are 0 while it's off.
Music imports are copies now, not hardlinks, because Arrmada writes clean tags (Settings →
Naming & metadata → Music naming & tags → Write house tags, on by default). While a torrent
seeds, its album takes up space twice. Turn Write house tags off to keep hardlinks.
A show's "Monitored" switch is now called "Searching: On / Off". Library scans add
shows with searching off (nothing is searched until you turn it on), and so does the Sonarr
import for shows Sonarr had unmonitored. Picking a "Monitor: …" option now also turns
searching on.
"Stalled" now means half an hour without progress, in the Queue, the Dashboard tile,
the Homepage widget's queue.stalled and Needs you. A torrent that's only briefly waiting
for peers counts as downloading. Time a download spends queued or paused in qBittorrent
no longer counts toward trying another release.
Only an admin can connect Plex now. Changing the Plex server or signing in with Plex
for the connection is admin-only, like Jellyfin. Managers keep watch monitoring and Test,
and see read-only Plex and Jellyfin cards.
Changing a connection's address needs its key typed again. For Plex, Jellyfin, Prowlarr,
indexers and download clients, a saved key, token or password is only sent back to the
address it was saved for. Change the address and the form asks for the secret again.
Switching an indexer to another kind also asks for it. If you set ARRMADA_JELLYFIN_API_KEY
in .env and haven't connected Jellyfin yet, type the key with the address the first time.
Scripts that use a session cookie must send Content-Type: application/json. Writes
(POST, PUT, PATCH, DELETE) made with a browser session are refused if they come from
another site, and a request body must be JSON. Personal API keys (X-Api-Key or
Authorization: Bearer) are unaffected.
Reverse proxies must send X-Forwarded-For (or Forwarded). Without it, every
visitor looks like they're on your home network and they all share one sign-in limit.
Caddy, Traefik and Nginx Proxy Manager send it already. Plain nginx needs
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;. Cloudflare's visitor
header is now trusted only where Cloudflare can have set it. Cloudflare Tunnel needs
nothing, and ARRMADA_TRUST_CF_HEADER=1 or 0 overrides the automatic rule
(Who a visitor is).
First-run setup only works from your home network. A fresh install that's reachable
from the internet can no longer be claimed by whoever finds it first. To set up from
outside, add ARRMADA_ALLOW_REMOTE_SETUP=1 to .env, run docker compose up -d, and
take the line out again afterwards. Installs that already have an admin aren't affected.
Staff who sign in with Plex or Jellyfin are asked for their two-factor code. That
applies to admins and managers with two-factor on (a trusted device still skips the code),
and "Require two-factor for admins" now covers these sign-ins too. Requesters are
unchanged.
Setting the audiobook password needs your Arrmada password, plus your two-factor code
if that's on. An account that only signs in with Plex or Jellyfin needs to have signed in
within the last 15 minutes. Changing your Arrmada password, or an admin's "Sign out
everywhere", now signs listening apps out too. They sign in again with the audiobook
password.
The bundled qBittorrent (Compose installs) is always locked with a password, even
when its card in Arrmada was removed or switched off. The one exception: it's left alone
while another qBittorrent of yours is switched on, and Settings → Status says so. New
installs no longer let home-network devices in without a password. Its own web page is
published on the server only (127.0.0.1) once its container is recreated. To keep it
open to your network, add ARRMADA_QBIT_WEBUI_BIND=0.0.0.0 to .env. The update script
prints a note until you choose
(qBittorrent's own page).
The bundled SABnzbd's page gets a login (user arrmada and a random password) if it
had none. Admins see it on SABnzbd's card in Settings → Connections. A login you set
yourself is never changed.
The database and backups are readable by Arrmada's user only. The data folder
becomes 0700, and the database and backups become 0600, at the next start. A new
.env is 0600, and the update script tightens an existing one that you own.
Archive extraction has limits. An archive may unpack to at most 1.5 times its size
plus 1 GiB, and hold at most 20000 entries. One that goes over is skipped and reported,
and the rest of the download still unpacks.
Rolling back to v0.3.3 or earlier needs ./update.sh --rollback --with-db. This
version adds database migrations (0334 to 0351: book edits, the music library, the seed
goal, films' other files and more), so v0.3.x refuses to start on the upgraded database. --with-db puts back the backup from before the update, which loses anything
recorded since. On Unraid, restore that backup before changing the Repository tag
(Updating).
Subtitle sync-check refuses unknown language codes (POST /api/v1/subtitles/sync-check
answers 400). It used to accept any tag.
Alerts and webhooks name people without their email. "New request", "Auto-approved
request" and "Problem reported" now carry the person's Plex or Jellyfin name, or the part
of their account name before the @. Check any n8n workflow that matched on that field.
Deleting a movie with its files now empties its own folder (#59). Its artwork,
.nfo, subtitles, theme music and similar files go to the recycle bin with the video, and
the empty folder is removed. This only happens when the folder is clearly the movie's own:
inside a Movies folder, named with the movie's title and year, not a link and not linked to
from beside it, holding no other
video, no file of a type Arrmada doesn't know and no file named for something else, and no
other film in your library has a file recorded there (even a missing one). Otherwise only
the movie's files and subtitles go, as before, and the delete dialog says why the folder
stays. Only the files the dialog listed can go: one that turns up after you opened it stays.
With the recycle bin off, the artwork and metadata are never deleted; they stay in the
folder. Deleting several movies at once, deleting through the API without the dialog's
list, and deleting one version or one file work as before.

New

Untagged audio tracks are identified (#90). Plenty of releases tag their audio und,
or nothing, so the language badges showed ? and a profile's Must audio language
couldn't judge the file. Where the local AI (whisper) is installed, Arrmada now listens to
every such track in your library once, in the background: three 30-second samples from a
quarter, half and three quarters through, and the language most of them hear wins. When
the samples don't agree, or whisper isn't sure, the track stays unknown rather than taking
a guess. Tracks with a real language tag are left alone, and so are commentary tracks
and tracks tagged zxx (no spoken language). A sample that can't be listened to means
the track is tried again on a later pass.
- What it hears is kept in Arrmada's database, never written into the file. The badges
  show it as detected (a dashed chip, "English (detected)").
- A detected language never makes Arrmada replace a file. It counts when it's a
  language your profile asks for: a file whose untagged track is heard as your Must
  language fits, and its badge is green. A language it only seems to lack stays unknown,
  as before: it's never upgraded, re-grabbed or counted in the save dialog for it, and
  its badge has no colour (as for an untagged track nobody has heard) rather than red or
  orange, nor is it counted among the season's misses.
- It runs one file at a time, only while no subtitle is being made, at idle priority, and
  newly imported files go first. It never runs whisper alongside a subtitle: when one
  is waiting, it stops after the current sample and finishes the file later. Switch it
  off under Settings → Subtitles → Local AI (whisper) → Identify untagged audio.
  It's on by default, but does nothing until a whisper model is installed.
- The AI subtitles' own check of an untagged track (before choosing transcribe or
  translate) now takes the same three-sample vote instead of one sample.
A film's cinema, digital and physical release dates (#93). The movie page and the
Discover sheet show when a film reached cinemas, digital stores and disc, as Radarr and
Overseerr do, and the Calendar and your calendar feed list a library film's digital and
physical releases as entries of their own (Digital release, Physical release). The
dates come from TMDB, for your region (Settings → Connections → Discovery region).
Where your region has no date for one of them, the US date is shown, then the earliest
anywhere, with the country beside it. Films you already have get them in the background
after you update, or straight away from Refresh & rescan. After you change the region,
they're read again in the background from the next restart (or from Refresh & rescan).
They're only shown: when a film is searched for still goes by its minimum availability and
its main release date, as before.
Other files in a film's folder. When a film's folder holds a video Arrmada doesn't use
(another cut, an old copy, a second film left there), the film's page shows +1 other
file by its file, and the Movies library has a Has other files filter and a small
+1 on the poster. A library scan or Refresh & rescan notes them, and the chip's
list looks again. It only says they're there and not tracked: nothing is moved, renamed,
recorded or deleted, and there's nothing to do with them from the list.
- Not counted: the film's own files and any other film's, samples (named so, in a Sample
  folder, or under 50 MB), Plex and Jellyfin extras (names ending -trailer,
  -featurette, -behindthescenes, -deleted, -interview, -scene, -short,
  -other, and folders such as Extras, Featurettes, Trailers and Deleted Scenes), and
  anything that isn't a video. Subfolders are looked in two levels deep, except one
  named as another film (Alien (1979)/Aliens (1986)).
- A folder that isn't plainly the film's own (not named for it, or another film in your
  library has a file there) is looked at only at its top level, and only files no film
  uses count. A film loose in a Movies folder has nothing listed.
Manual import can look anywhere you gave Arrmada, not just the downloads folder. On a
movie, show or book, Manual import now opens with the finished downloads whose names
match it (in your download clients, Not managed ones included, or sitting in a downloads
folder), one click to use. Choose a folder… browses your downloads and library folders
with how many videos (or book files) each folder holds and a filter, searches every
downloads folder by name, or takes a typed path. It only ever shows folders inside your
downloads and library folders, never Arrmada's own data folder, and doesn't follow links
that lead outside them.
A global seed goal (#78). Settings → Downloads → Seeding sets a ratio and a
seeding time for indexers set to Use the global default, so you don't have to type the
same limits on each indexer. It only ever applies going forward, and only where you choose
it. Every indexer you already have keeps its own goal, and one set to seed forever (both
at 0) still does: switch an indexer in its Seed rules (Global default, Own goal or
Seed forever) to use the global goal. An indexer added while the goal includes a
time, by hand or from Prowlarr, starts on it; otherwise a new one still gets 14 days. A
ratio-only goal is never given to anything automatically, because it could end a private
tracker's torrent before its minimum seeding time; choose it for an indexer and it
applies. Prowlarr rows you already have aren't changed.
- Torrents grabbed from an indexer on the global default follow it as it is now, so
  raising, lowering or clearing it reaches them. With it cleared, they seed for the
  standard 14 days. Torrents grabbed before this release keep the goal they were grabbed
  with. A torrent you add by hand or upload follows the global goal while it includes a
  time, and gets the standard 14 days otherwise.
- A torrent is removed at whichever limit it reaches first, so a ratio can end a seed
  before its time is up. For a private tracker's minimum seeding time, set a time goal
  alone. Removal works as an indexer's goal always has: only once Arrmada has imported the
  torrent, with its data, since the library keeps its copy; a torrent you dismissed in
  Review is removed with its files kept.
- The fields save only when you press Save. Check finished torrents shows what the
  typed goal would do before you save it. For each finished torrent it lists the ratio and
  seeding time against the goal that applies, whether that goal is the live global default
  or one recorded when it was grabbed, and what the next check will do. It only reads;
  nothing on it removes anything. It pairs older downloads with their grabs the same way
  the next check will, without saving the pairing.
Seed cleanup removes less (#78).
- At most 10 torrents go per check (every 10 minutes), whatever ends them — a goal, the
  global goal, or seeding switched off for the indexer. The rest wait for the next check,
  and the log says so.
- A torrent is matched to its grab by its hash only, so a cross-seed or a copy you added
  with the same name is no longer removed with its data. A download grabbed before hashes
  were kept is matched by name only when no other download in the client has that name,
  and one grab never removes two torrents.
- A season pack whose import left files it couldn't match to an episode keeps seeding
  instead of being removed with those files, whether it was imported automatically, from
  Review or by a manual folder import.
- An indexer edit checks its seed values only when they change, so an old goal over the
  new five-year limit doesn't block other edits.
A Missing episodes filter on Series (#82): shows with aired, monitored episodes that
have no file, whether searching is on or off.
Put misplaced files back (#56). Episode files that v0.3.0 and v0.3.1 imported into a
folder named like a season right inside the TV folder (<TV folder>/S13/Season 12/…) are
listed on their show's page, with a preview of where each goes. Nothing is ever replaced:
where the show's folder already has the episode, the misplaced copy goes to the recycle
bin only when it's identical or lower quality, and otherwise both are left for you.
Emptied folders are removed, and the show's History records what moved. Settings →
Status lists every show affected.
Built-in SABnzbd and Prowlarr in the Unraid container, the same ones the Compose
install adds with --with-sabnzbd and --with-prowlarr. Both are off unless you set
Built-in SABnzbd or Built-in Prowlarr to true under Show more settings (a blank
field counts as off). SABnzbd is connected to Arrmada by itself, with Arrmada's categories,
and saves to usenet/ in your Downloads folder so imports stay on the same drive; you add
your Usenet provider on its page (port 8085). Switch it off again and its client in
Settings → Connections is switched off with it. Prowlarr's page is on port 9697: add your
indexers there, then its API key in Arrmada; it gets the built-in FlareSolverr like the
Compose install's Prowlarr. Their settings live in Arrmada's appdata. The container image
is about 220 MB larger on disk (about 85 MB more to download) whether or not you use them.
Already installed from the template? Unraid keeps your old template, so add the switch and
port lines yourself: Arrmada on Unraid lists them. The docs no
longer suggest SABnzbd or Prowlarr were inside before. Thanks to @aleonenko-us, whose PR #66
started this.
KEPUB for Kobo. EPUBs can be downloaded as KEPUB from My shelf and the book page
("Download for Kobo (KEPUB)"), and the OPDS catalogue offers a KEPUB link beside each
EPUB. The KEPUB is made on download from a copy (the library file is never changed) and
cached. If the conversion fails, you get the EPUB. A switch in Settings → Book server turns it off. The e-reader panel gains a
"Kobo (stock reader)" row.
Calibre folder scan. The library scan reads Calibre's metadata.opf to tell which
book a folder holds: its ISBN first, then its title and authors. Series and publication
date fill gaps the catalogue leaves, and cover.jpg is used only when the catalogue has
no cover. Calibre's " (123)" is taken off a folder name only when the folder really is
Calibre's, so your own "Saga (2)" keeps its number. Nothing is written to the Calibre
folder.
Edit a book's details (book page → Fix → Edit details…): title, subtitle, author(s),
series and position, description, year, language, tags, ISBN, narrator(s) and publisher.
Each edited detail is marked, shows the catalogue's value, and has Reset to matched (or
Reset all). Refreshes, the Hardcover upgrade and merges never undo an edit. Change match
asks whether to keep your edits. Edits show everywhere: Books, ⌘K search, series,
requests, listening apps and OPDS. Files are never renamed by an edit. When an edit would
change the files' names, the page offers Preview rename for that book.
The disk guard can measure a different drive. Settings → Downloads → Download disk
guard → "Measure a different drive". On Unraid, the downloads share reports the whole
array while torrents fill the cache pool. Map the pool read-only (for example /cache)
and the guard watches that. A new "Also pause when free space drops below (GB)" limit
helps on ZFS pools, where the percentage counts only the root dataset. A drive that can't
be read, or that turns out to be a RAM folder (a mistyped Unraid pool name), is a clear
warning, and nothing is paused or released on its account
(Download disk guard).
Staff choose a quality profile when requesting (#50). On a Discover movie or show,
admins and managers get a split Request button ("Request · HD 1080p") with the profiles
for that kind of media; for a show it's on the season picker's confirm button. Leave it
alone and the default is used. Requesters see no change, and a profile a requester sends
through the API is ignored (staff still choose when approving).
Order the library your way (#36). Movies and TV list views get sortable Year, Added and
Rating columns (click a header, click again to reverse), the Sort menu and the headers share
one remembered order, and a new ↑/↓ button reverses it in grid view and on phones. The Books
table sorts from its headers too, with a new Added column.
Browse one library folder (#34). With several Movies or TV folders, the library page gets
a Folder picker (?folder= in the address, so a filtered view can be bookmarked). Search,
counts and Select all stay inside the chosen folder, and Add from a filtered view puts the
new title in that folder.
Descriptions in your language (#35). With Settings → Library → "Show titles as" set to
My language, movie and show descriptions on the title pages and in Discover come in that
language when TMDB has a translation, and in English otherwise. English and Original modes
keep English descriptions; season and episode descriptions stay English. Existing libraries
catch up in the background, about one title every two seconds.
Rename fills in what the file name lost (#48). The Rename preview for movies and episodes
now takes quality, source, codec and release group from the release Arrmada recorded, so a
naming scheme that gains {codec} or {group} works for files named by an older, shorter
scheme. Safeguards: a release that contradicts the file name, or belongs to a file replaced
outside Arrmada, is ignored; the codec the file actually has wins (Convert); editions never
come from the release; renames after a TV renumber still use the file name only. Nothing is
renamed until you apply the preview.
A Stalled filter on Activity, and which client each download is in. The Queue gets a
Stalled pill (no data for half an hour while trying to download, the same rule Arrmada
uses before trying another release) with a "Try another release" button, and the
Dashboard, Homepage widget and Needs you now count stalled the same way. With more than one
download client, each row says which client it's in, with a client filter.
"Searching: On / Off" for shows (#52, #53). The show-level switch that decides whether
Arrmada searches for a show now says so, and "Monitor" only means which episodes you want.
Wanted episodes of a show with searching off say "Searching off" instead of "Unmonitored",
with a banner to turn it on. Picking a Monitor option turns searching on and the menu
remembers it (or shows Custom). The Series list counts Searching on / off, and turning
searching on for many shows at once is searched a few at a time, like an import.
SECURITY.md: how to report a vulnerability privately, and which versions get fixes.
The README now explains how Arrmada is built and tested.
Music (Preview) on the Dashboard and Calendar. With Music on, staff get an Albums tile and a
Music card (library, missing and upcoming albums, new releases, recent imports), album release
dates on the Calendar and in their personal calendar feed, and music counts in the Homepage
widget. Requesters don't see albums.
Music (Preview) keeps up with new releases. Every few hours Arrmada checks your artists
(through the cache and MusicBrainz's polite limit); a genuinely new album from an artist set to
follow new releases is wanted and searched once, and listed under New releases on the Music page.
Release types work properly ("Album + Live" means live albums), singles already on an album
aren't wanted, and "Latest" never picks an unreleased album. Per-artist Options on the artist
page, a library default in Settings. Nothing old is ever wanted by surprise, changing settings
never wants anything by itself ("Want matching albums…" shows a count first), and albums
MusicBrainz removes are kept, files and all, but not searched automatically.
Music (Preview) splits single-file albums. A rip that's one big FLAC, WAV or WavPack image
plus a .cue sheet is matched first, then cut into one FLAC per track at the exact sample
positions (gaps stay where CD players put them), proven sample-exact against the image, and
imported like any other album. It streams, so even a 24/192 image needs well under 100 MB of
memory. A cue that doesn't fit the image, or the album's length on MusicBrainz, goes to Reviews.
Music (Preview) writes proper tags and follows your naming. Each import is copied into a
hidden staging folder, given clean tags (every artist, MusicBrainz IDs, compilation flag,
embedded cover plus cover.jpg), checked so the audio is unchanged, then moved into place in one
step; your downloads are never written to, and a crash at any point recovers on the next start.
Settings → Naming & metadata has music naming templates with a live preview, and albums get
Rename files… (preview first) and Write house tags. Artwork, booklets, cue sheets, logs and
lyrics move with the album.
Music (Preview) matches downloads like a person would. Every edition of an album is known,
with the earliest official release picked by default and a Change edition… picker. Downloads are
scored against each edition (the beets method), partial albums import what fits, and close calls
go to Reviews with the reason in words ("length off by 45 s") and an Import as option.
Thresholds are in Settings → Naming & metadata → Music matching. Badly tagged files can be
identified by sound with your own free AcoustID key (Settings → Connections).
Music (Preview) has a new library model: albums have editions, discs and credited artists,
and each track file keeps its measured quality. Your existing music moves over on update with
nothing lost; anything that can't map one-to-one (two tracks claiming one file) shows as a note on
the album page with a Resolve button, and that album isn't searched automatically until then.
Music (Preview) is safer while it's rebuilt. Artist and album names fold properly
(Björk, AC/DC, R.E.M., "The The") without matching the wrong band ("Fall Out Boy" is no
longer The Fall); words like "Escape" no longer read as lossless; quality is measured from
the files; track and disc numbers are read safely; an upgrade never lowers quality and old
files only ever go to the recycle bin; files that match no track are left in the download
instead of being cleaned up; albums show as complete again. MusicBrainz lookups share one
polite limit and a permanent local cache, so music keeps working when MusicBrainz is down, and
Settings → Connections → Metadata can point at your own MusicBrainz copy.
Music (Preview) searches and upgrades on its own terms. Album and artist pages say why an
album isn't downloading (last outcome, next try) and have Search now; an artist's Search now
takes up to 25 albums a click, never-searched first, and a second click moves on to the rest.
Interactive search lists every release with the reason it would or wouldn't be grabbed, and a
manual grab imports into the album you picked (the same release twice is refused while it's
still downloading). Video and implausibly sized releases are dropped. With upgrades turned on
for a profile, a paced sweep replaces albums with strictly better quality, never a downgrade,
old files to the recycle bin.
Music (Preview) scans big libraries fast and keeps an eye on your files. The library scan
reads tags (files with MusicBrainz IDs match straight away), runs in the background, and a
rescan only opens folders that changed. It never moves, retags or deletes anything; folders it
can't place go to Reviews with Match…. Albums moved or retagged outside Arrmada (Picard,
beets) are linked again at their new place. A daily "Check music files" task notices files that
vanished or changed outside Arrmada and shows them as gone from disk; it never deletes anything
and never downloads them again by itself. Artists are found under their other names too
("Beyonce", "宇多田ヒカル").
Actors on a library page open their Discover page (#55). Click or tap someone in a
movie's or show's Cast row to see the other movies and shows they're in; Close or Back
returns to the library page. Titles added before this update link their actors after
their next metadata refresh (Refresh & rescan on the page; monitored shows also
refresh on their own schedule).
Add the year to a show folder only when it's needed (Settings → Naming → Series naming),
the way Sonarr's TheTVDB names look: "Breaking Bad", but "Dark Matter (2024)" because another
Dark Matter exists. The year is kept when another show in the library has the same title, a
folder with that title is already in a TV folder, or TMDB lists another show with that title
(titles compare ignoring case, punctuation and a leading "The"). If a TV folder or TMDB can't
be checked, the year is kept. It only names new show folders, once per download, so a
download never splits across two folders. A show that already has a folder keeps it, and
Rename names files inside that folder (it only gathers a show that's split over two folders
into the one it uses). It's off by default, so nothing changes until you turn it on.
A rename never gathers a show into another show's folder. A move into a folder where
another show keeps its files is refused and shown as a conflict in the preview. Season
folders named "S01", "Staffel 1" or "Saison 1" are recognised as part of the show.
A guide to downloading through a VPN (#92,
docs/vpn.md): qBittorrent
behind Gluetun with a docker-compose.override.yml, keeping Arrmada and FlareSolverr on your
normal connection, and what to do on the Unraid all-in-one container. Arrmada doesn't test it.

Changed

"Released" now waits for the digital or disc release (the cinema date + 90 days when
neither is known); pick In cinemas for cinema-time grabs. The earliest such release in
any country counts, so a film out on US digital isn't held back by a later local date. Before, "Released" counted
from the cinema release, when almost every release is a cam. The scheduled search, RSS,
search on add and a request approval's search all wait; Search on a film doesn't. The
film's page says what it waits for ("Waiting for the digital release on 12 Dec"), and
Wanted, the Calendar ("search starts"), Discover and request cards show the same day;
approving one says "searches from 12 Dec". A cinema re-run or a new 4K disc never holds
an older film back, and the dates of films still waiting are read again daily, so a
digital release TMDB learns of later counts.
A library scan no longer guesses which show a year-less folder is when more than one
show has exactly its title ("Dark Matter": 2015 and 2024). The folder goes to review with the
candidates, so you pick. A folder with a year, or a title only one show has, matches as
before.
Built with Go 1.26.10, because Go 1.25 is out of support. The Docker build pins Go to
the patch release. Building from source outside Docker needs Go 1.26.
Vulnerability scanning in CI. Every change runs govulncheck (it fails when our code
can reach a known-vulnerable function) and npm audit (high or critical, in the UI's
production dependencies). CodeQL code scanning runs on Go and TypeScript. Go modules with
known advisories were upgraded (golang.org/x/net, x/crypto, x/sys,
golang-jwt/jwt).
Sign-in limits are shared. The audiobook server and the OPDS catalogue count wrong
passwords with the main sign-in limiter (10 per address in 15 minutes, then a pause per
username from outside your network). Your own devices that signed in recently aren't held
up, and an e-reader repeating the same stale password counts once.
Web Push only sends to the browsers' own push services, and an account keeps at most
10 devices (the oldest goes).
The image's Apprise is pinned and installed with hashes.
Release notes on GitHub now come from this changelog, with a Before you update section
for each version (#31).

Fixed

Two shows with the same name no longer get each other's episodes (#81). With Scrubs
(2001) and the 2026 Scrubs both in the library, a release named Scrubs.S02E03… doesn't
say which show it's from. A search for one show grabbed it, and the import trusted the
grab, so the revival's S02E03 could replace the original's. This only changes anything
for a show that shares its name with another show in your library. Releases that name
the year (Scrubs.2026.S02E03…) or the country (The.Office.US…) work as before. For a
release that names neither:
- Searches use the release's publish date. A release posted before a show's season
  began can't be from that season, so Scrubs.S02E03 posted in 2015 is grabbed for the 2001 show
  and never for the revival. A release for a season the other show doesn't list yet
  only counts as the older show's if it was posted before the other show first aired.
- When a release could belong to either show, or the indexer gave no date, the
  automatic searches, upgrades and quick Grab buttons skip it. The show's last
  search says so, and Choose release still lists it with the reason, so you can
  grab it by hand. In practice, the newer show only grabs releases that name its year.
- A grabbed download that could belong to either show goes to Needs you as Which
  show?, naming both shows. It isn't imported or used to replace a file. Import
  anyway puts it in the show it was grabbed for; Choose different show… puts it in
  the other one. Either way, it only replaces a file it's better than; when it isn't
  better, Needs you says Kept the existing file and leaves the download as it is.
- A release you pick by hand from Choose release imports as before.
Library episodes named "1x8" are read (reported in #81). A show whose files are
named with a single-digit episode — "11.22.63 - 1x8- The Day in Question [HDTVMux1080p
Ita 5.1].mkv" in an "S01" folder — had none of them read: the "1x08" form needed a
two-digit episode so an aspect ratio ("16x9") or a dimension ("4x4") is never taken for
one, and Refresh & rescan found 0 of 8 episodes. The library scan, a rescan and the
duplicate check now read a single-digit "NxE" in a file already in a show's folder when
it can't be anything else: after the show's title and set off by dashes ("Title - 1x8 -
Episode", "Title - 1x8- Episode", "Title – 1x8"), or after the title inside the season
folder of the same season ("S01", "Season 1", "Stagione 1"…). Never inside brackets or
after the release tags; never at the start of the name, where it is usually the show's
own title ("3x3 Eyes 05"); never "0x5" outside a Specials or Season 0 folder; never
against a season folder saying another season; never two candidates ("Kids 2x2 - 2x5");
never "1x8x9" or "1x8-9", whose other episodes can't be told; and "16x9", "4x3", "4x4",
"2x4" or the same number twice ("3x3", "1x1") only when both dashed and in that season's
folder. A file read this way only fills an episode nothing else holds: it never replaces
an episode's recorded file or another copy read by its name, and stays on disk listed as
a duplicate instead. Release names (searches, grabs, download imports) and Manual
import read exactly as before.
A rescan no longer swaps an episode's file for one that isn't named as the show.
Another show's episode dropped in a show's season folder (Beta - 1x08 - Other.mkv in
Show (2010)/Season 1), or a file with no title in its name (S01E08.mkv), could take
the place of the episode's recorded file on Refresh & rescan when it was the bigger
or sharper copy, and the duplicate check then offered the show's real file for deletion.
Now:
- A file whose name isn't the show's never replaces the file an episode has. It still
  fills an episode with no file in the show's own Title (Year)/Season N folder, as
  before, so episodes named in another language or in romaji keep being found; when an
  import later brings the episode's file, the filled one goes to the recycle bin like any
  replaced file. When several files could fill it, the one named as the show is taken
  first, then one with no title, then one named otherwise, and only then the better copy.
- A copy of the show's own episode whose name doesn't match the show (a romaji title the
  show has no alias for, a rip with no title) is no longer swapped in on rescan for being
  the better copy, and isn't listed as a duplicate either. Import it by hand, or add the
  name under the show's Alternate titles and rescan.
- The duplicate check never offers the show's own copy for deletion beside a file named
  otherwise, and never offers a file named otherwise. It never offers a two-episode file
  while it is the other episode's only copy: when that episode has no file recorded, or
  its recorded file is gone, deleting it would have removed that episode too (this
  happened in v0.3.3 as well).
- A file recorded under a path that differs only in letter case is still the episode's
  recorded file on a disk that ignores case.
- Episodes an earlier rescan already switched to another show's file are left as they
  are: check the duplicate list and the episode's file, and import the right one by hand.
A season pack is no longer blocklisted and grabbed again while qBittorrent moves it
(reported in #94 by cpetzel). With qBittorrent's Default Download Path (incomplete
torrents) on, a finished torrent is moved from that folder to its save folder, and
until the move is done its path still names the incomplete folder, which Arrmada usually
doesn't mount. A TV import that ran during the move found no folder, took that for a
release with no video, blocklisted it and failed its grab, and the next search grabbed
another copy of the same season. Now:
- Imports (films, TV, books and music) wait while qBittorrent is moving a torrent, or
  says its files are missing, and pick it up from its final folder once it's there.
  Import suggestions, unmanaged downloads and Check paths wait for the move too.
- A TV download whose folder Arrmada can't read is tried again every sweep. It is never
  blocklisted, and its grab stays as it was. If it's still unreadable after five
  minutes it shows once in Needs you with the folder and the error. A season pack
  that was already imported and is only being re-read for the season's gaps is just
  skipped. Only a folder that was read and holds no video is still treated as junk.
- A download with a folder inside it that Arrmada can't open (the wrong owner, say) is
  held, and shown in Needs you, rather than partly imported. It is never judged to
  hold no video, so it is never blocklisted or deleted for it. The folder it couldn't
  open may hold the episodes, or the film itself beside a featurette that would have
  been taken in its place. A file that disappears while the folder is read (the
  download being moved, rechecked or renamed) means the folder is changing: nothing is
  decided, and it's read again on the next sweep.
- Importing such a download from Review says Arrmada can't read the folder, instead of
  "no video files were found". Choosing files by hand for a show lists what could be
  read and says part of the folder couldn't be.
- A torrent qBittorrent has been moving for over two hours shows in Needs you as a
  move that's taking too long, since it won't import until the move finishes. It isn't
  counted as a stalled download and doesn't send a "stalled" alert.
A release already blocklisted this way stays blocklisted. Remove it from the show's
blocklist, and if its torrent is still in qBittorrent the next import picks it up.
Convert and imports no longer remove a video Arrmada doesn't track.
- Converting Heat (1995).mp4 writes Heat (1995).mkv. A file you kept under that name
  (a remux beside the film, say) used to be moved into the originals hold with the
  original, and deleted with it when the hold expired. Now the conversion waits under
  Convert's Problems until you rename or remove that file, and nothing is encoded until
  then. An empty file there is still moved aside, as before.
- A film import whose name was taken by a file no record owns used to send that file to
  the recycle bin, or overwrite it with the bin off. Now the file stays, and the import
  lands beside it with the version's label (Heat (1995) - 1080p BluRay - Default.mkv).
  An upgrade still replaces the film's own file under the same name, through the bin,
  and re-importing the same file still does nothing.
- A file earlier versions moved into the originals hold beside a converted film is no
  longer deleted when the hold is released. It goes back to the film's folder. If its
  name there is taken (usually by the converted file), it goes back as
  <name> (kept by Convert).mkv, and Convert's log says so.
- A revert that fails part-way and can't put its files back keeps both copies, and
  Arrmada tries again within the hour. Releasing the hold in the meantime used to delete
  both copies.
- An import that records a file already on record under another spelling (another case
  on a case-insensitive disk, or a symlinked folder) no longer recycles the file it has
  just imported. This applies to films and episodes.
- On a case-insensitive disk, converting Heat (1995).MKV no longer waits forever because
  Heat (1995).mkv, which is the same file, seemed taken, and the conversion can be
  reverted.
A show's search no longer picks a release you avoid over good season packs. A
release under your bitrate floor, or with a format, language or Unwanted item you avoid,
is meant to be a last resort. For shows it wasn't: the search took the release that
covered the most, so a starved complete-series pack (14.4 GB for five seasons, under a
5 Mbps floor) beat good season packs for every season. Now the search fills everything
it can from releases you don't avoid first — complete packs, season packs, then single
episodes — and only turns to an avoided release for what's still missing. If that is
the only release that covers a gap it's still grabbed. A torrent with no seeders is
likewise only used for what nothing seeded covers. An oversized pack is still the last
resort after all of that, so one missing episode is never filled with a whole series
when a single episode exists, avoided or not. An avoided pack that would mostly bring
episodes a better release is downloading right now waits until that download finishes,
so the better files land first; then it fills the rest. The import normally keeps the
better files over its copies. A download that stops moving, or is still going a day
after it was grabbed, holds nothing back. This applies to the missing sweep, the Search
and Grab buttons and the request preview; a Grab that leaves a pack waiting says so.
A request's release preview searches the seasons asked for by number, as the
missing sweep does: up to six, plus the broad search. It searched only the first, so a
show whose season packs only turn up when asked for by season was previewed — and, on
Approve & grab, grabbed — with a worse pack instead. The note under it says what was
found for the other seasons, and which weren't checked.
A request's release preview names every season a pack covers. A complete-series
pack showed as "Season 1: 1 pack". It now reads "Seasons 1–5: 1 pack". When the pick is
a release you avoid, Why says so plainly: that nothing better was found, or that
better releases were found but none covered everything asked for, and what it falls
short on.
A season pack whose episode titles disagree with the show's episode list waits in Needs
you instead of landing on the wrong episodes. A show's episode list can be out of date
when a whole season is released at once. Moving a file to the episode its title names
then put it on an episode another file in the same pack was numbered for, and that file
was skipped as already imported. Two episodes ended up with the wrong file and two with
none. Now Arrmada first refreshes the show's episode list, and if that settles it the
pack imports as normal. If it doesn't, nothing from the pack is imported. Needs you shows
it as Titles disagree, names the files that disagree and what each one's number and
title say, and Map files… lets you place them by hand, starting from their
filenames. Packs whose whole numbering is shifted from the episode list (Parks and
Recreation's two-part "London"), and a REPACK or PROPER beside the file it replaces,
still import by title as before.
Dismissing a held season pack sticks. If the season still had missing episodes, the
pack was checked again, held again and alerted on again.
Grabbing an already-imported release again by hand imports it again. It used to be
added to the download client and then ignored, because Arrmada had recorded it as
imported and often blocklisted it for the show. The new grab now clears both, so the
release imports like any release you pick yourself. The earlier grab no longer counts
for seed cleanup, so if you dismiss the new grab in Needs you, its files are kept when
the torrent is removed at its seed goal. Files the earlier import couldn't match still
keep the torrent seeding.
Map files… keeps a download seeding while files are left unmapped. Files you leave
out exist only in the download, so the torrent is no longer removed with its data at its
seed goal.
The Activity search shows only what matches (#71). With the same qBittorrent added
twice, every torrent was listed once per client under one hash, and a search left
unrelated cards on screen and added copies each time you retyped. Cards are now told
apart by client, and Needs you lists such a torrent once.
A download never reads 100% before it's finished (#73). Below 100% the queue, a
film's page, posters and episode chips show one decimal (99.6%), and request cards round
down.
Built with Go 1.26.10, which fixes a Go security advisory in certificate checking
(GO-2026-6809), and with golang.org/x/net 0.62.0 (GO-2026-6860).
With two download clients, the disk guard only pauses the full drive's downloads
(#70). A download placed on a drive of its own client's (its folder as Arrmada sees it,
through the client's path mappings, or SABnzbd's incomplete folder for a job still
downloading) is judged by that drive, so a second client on a volume of its own keeps
downloading while the first one's drive is full, and the other way round. Anything the
guard can't place is judged by the Downloads folder's drive, as before, and named once in
the log. Settings → Downloads lists each client's drive with the guard's reading. With one
download client switched on, nothing changes.
Episode thumbnails from TVDB load again (#76). TVDB can give an episode image as a path
rather than a full address, and the browser then asked Arrmada for it ("Page not found").
New and refreshed episodes get TVDB's artwork address, and the update fixes the ones
already stored.
Shows with S01-style season folders work fully again (#56). A show stored as
<TV folder>/Show/S12/ (an S{season00} season folder template, Stagione 1, Saison 1…)
is read as living in Show, as v0.3.2 and v0.3.3 couldn't: new episodes go into the
season folder the show already has for that season, whatever it's called (never a new
Season 1 beside a Stagione 1; while that folder is unreachable, the download waits
instead), a new season gets its own folder inside Show named by your template, and
Refresh & rescan, Rename, the duplicate check and Move to folder work again.
Shows v0.3.2 and v0.3.3 refused for their season folders' names read normally after the
update, with nothing to do. One whose folder also holds another show — another show's
folder, a video loose in it that isn't the show's, a link — still waits in Needs you, and
Move to folder refuses it. If you started a show over as v0.3.2's notes describe,
nothing changes for it.
Shows kept inside a category folder stay there (#56). A show in
<TV folder>/Anime/Show/ (or under a TV folder set one level too high) gets its new
episodes in its own folder, S01-style season folders included, where v0.3.1 put them in
a new <TV folder>/Show/ beside the category. A show's folder is never a folder above the
one holding its files (other than the show's folder above its season folders), never a
folder named like a download or disc unless it's right inside the TV folder, and never a
folder that also holds another show, tracked or not; when it can't be told, the download
waits in Needs you.
Refresh & rescan, the duplicate check and the library scan read only the show's own
files (#56). They read exactly the same files: those right in the show's folder, in its
season folders (and disc or download folders inside them), or in its own download
folders. They never read a folder inside it that holds another show, whether hidden,
nested deep, or under a folder called Other or Bonus. In a show folder right in a TV
folder that's provably the show's own, an episode right in it or in a Season N folder is
the show's however it's named (Shingeki no Kyojin, Frieren,
Top Gear 2023 S33E01) unless its name says it's another show in your library
(Bluey - S01E02.mkv, or The Office (2001) - … beside the 2005 show) — that one is never
claimed, nor offered as a duplicate to delete. Elsewhere (S01-style season folders, a
show inside a category) a file's name must fit the show. The duplicate check offers for deletion
only files the show has recorded or whose name says they're the show's. A file whose
name has no title
(S01E01.mkv, 01.mkv) counts only in a folder where the show already has files. A
video loose in the folder of a show kept in season folders isn't taken for an episode.
Featurettes in Plex extras folders (Extras, Featurettes, Behind The Scenes, Deleted Scenes,
Interviews, Scenes, Shorts, Trailers, Other) are never taken for episodes. Files the show
already has always count. A show with no files yet reads only its own Title (Year)
folder right in a TV folder, and only when that's provably its own — never a year-less
Title, which may be a namesake's. Adding a show with Scan library marks only the
files a rescan would read. A theme song, trailer or sample loose in a show's folder, and a
media server's backdrops or theme-music folder, are ignored.
A link inside a show's folder keeps it from being taken for the show's alone (#56). A
link to another folder or to a video means the download waits in Needs you, and Move to
folder won't move the folder.
A show whose folder can't be told has a way out (#56). Move its files by hand out of
the folder no show should own (one named like a season, one other shows are in, the TV
folder itself) into a folder named with the show's year (Show (2019)), then use
Refresh & rescan. Each such file is found by its name in that TV folder, as long as the
name has the show's title and an episode number, exactly one file has it, and no other
show has that file. A folder without the year may be a namesake's, so nothing is followed
into it (unless the show's other files are already there). Nothing is changed unless the
show then reads as living in one sound folder holding every file found; otherwise the
rescan is refused and the show's records stay exactly as they were. A file gone from the
show's own folder is never looked for elsewhere (it's deleted, or on a share that's
away), and nothing is followed while the TV folder looks unmounted. Deleting the show in
Arrmada keeping its files and using Scan library works too.
Movies: deleting a movie left its artwork and .nfo files behind, so its folder stayed
(#59). The delete dialog now lists every file that goes and whether the folder is removed.
If the recycle bin won't take one of those extra files, it stays and the delete still
finishes.
Movies: deleting a movie or its file no longer removes a folder that is a link to another
folder. Before, the link itself could be removed, however much was in the folder it
pointed to.
Movies: deleting a movie no longer takes a subtitle named for another film in the same
folder whose video is missing ("Movie.Directors.Cut.en.srt" beside "Movie.mkv").
Movies: deleting a movie on a large library over a network share no longer takes tens of
seconds, which could make the delete dialog time out behind a reverse proxy.
Movies: Move to folder could take a whole folder that wasn't the film's own. A film
sitting loose in a folder such as Movies/Marvel took everything in Marvel with it,
including films Arrmada doesn't track and the folder's artwork. Onto another drive, those
originals then went to the recycle bin. A film's folder now moves only when it's clearly
the film's own: named for the film, not reached through a link, with nothing in it but the
film's files, its subtitles, and artwork or .nfo files named for it. If the folder holds
anything else, the move is refused and the dialog says why. Use Organize to give the film
a folder of its own first. The Move dialog now lists every file that will move before you
press Move, and the move never takes more than that list. TV shows move as before.
Audio languages: a profile with Italian as Must no longer grabs MULTi or Dual Audio releases
that don't name Italian (#58). Release names that write FRA, DEU, NLD, ESP, JAP or VFB (in
capitals, outside an episode title) are read as French, German, Dutch, Spanish, Japanese
and French.
Movies: renaming a movie (Rename files or Organize) took the artwork from the folder it
sat in, even when that folder wasn't the movie's own. A film sitting loose in a franchise
or collection folder carried the collection's folder.jpg and poster.jpg off with it.
Now a movie takes its artwork and .nfo files only from a folder that is clearly its
own. That means a folder named for the movie, holding no other video, with no other
film's file recorded in it, no links, and only files Arrmada recognises. Files named for
the old file name get the new name. They go only into a new folder, or one that holds no
other film: with a naming scheme like "{title}", Heat (1995)'s poster no longer lands in
the "Heat" folder of Heat (1986). Anything else stays where it is, and the rename
preview lists every file that moves and says why the others stay. A subtitle that
belongs to another film's file is no longer taken along. A folder that is a link, or one
of your library or downloads folders, is never removed.
Library folders: a show, artist or album whose name is only dots can no longer reach
outside its library folder. Names like "..." are still allowed.
Archive extraction: a refused archive no longer stops the rest of the download. Zips of RAR
volumes and subtitles nested in zips unpack.
Books: a merge keeps edits honest, and a title edited into a different one ("Dune" →
"Dune Messiah") no longer picks up the old book's releases.
Books: the Calibre scan no longer replaces a cover or series set while it was running.
Symlinked or odd metadata.opf files are ignored.
Books: the Kobo download shows for a Calibre folder that holds an AZW3 and an EPUB.
Managers no longer get a Jellyfin editor that fails on save.
CodeQL fixes: the all-in-one image refuses a PUID/PGID past 32 bits instead of
running as root, and file names are built only from checked values.

Security

The fixes listed under Before you update: CSRF protection, client address and
Cloudflare header, LAN-only first-run setup, two-factor on staff Plex and Jellyfin
sign-in, the audiobook password step-up, saved secrets kept with their address,
admin-only Plex connection, locked bundled qBittorrent and SABnzbd, private database and
backups, and archive limits.
Cover images: a requester-supplied cover is kept only from a catalogue image host. The
image cache fetches only public addresses and real raster images (no SVG), and serves
them sandboxed. Before, a book request's cover could carry a script or reach into your
network.
Jellyfin: saving a new server address with the API key left blank sent the saved key to
the new address. It's refused now, and the form asks for the key again, as the other
connections do.
Keys and tokens sent in headers never follow a redirect to another server.
The Plex token only goes to servers that say they belong to the account, and a manager can
no longer repoint Plex sign-in at their own server.
Sign-in checks take turns, so a burst of guesses can't slip past the limit.
.env is 0600 on new installs, and edits keep its mode.
The release workflow's third-party action is pinned to a commit, and CI's token is
read-only.

Full Changelog: v0.3.1...v0.4.0