v1.33.1

A series fill could grab your whole library with auto-grab switched off.

Ten fixes for faults reported within a day of v1.33.0, most of them found by
one person exercising Hardcover as the primary metadata provider harder than it
had been exercised before. Four are in series fill, which turned out to be the
one action that ignored the auto-grab kill switch, could add the wrong book,
created the box sets the previous release taught the catalogue to remove, and
never recorded the Hardcover link it had just resolved.

Fixed

• Auto-grab off now stops every grab, not most of them (#2256). The kill switch was checked by each caller rather than at the point a download is dispatched, so three paths never consulted it at all: a bulk Search action, adding a single book with search on add, and adding a book from Recommendations. All three fanned out indexer searches and grabbed releases with Auto-grab switched off. The check now lives inside the scheduler's search and grab entry point, so it holds for every caller including any added later, and a test fails the build if a new dispatch site routes around it. Books are still created and still marked wanted and monitored, so nothing is lost from the Wanted page. Only the automatic download stops.
• Series fill ignored the auto-grab kill switch (#2242). With Auto-grab off in Settings, a fill still fanned out indexer searches and grabbed the whole series. One report had fourteen releases grabbed and imported a minute after the switch was turned off, seven of them already in the library. Fill now creates the books and marks them wanted and monitored but queues no searches, which is what the scheduled wanted sweep already did.
• Series fill added the wrong book (#2238). When a Hardcover series files several books at one position, a box set, a novella and the real volume all at position 1, clicking add on one of them created whichever Hardcover happened to list first. An explicit book id now wins over a position match.
• Series fill created box sets, and an excluded box set blocked the real book (#2239). v1.33.0 taught the author catalogue to drop titles that unambiguously name a bundle, but series fill created books on a different path that consulted no filter at all, so the box sets came straight back in through it. Fill now applies the same check. An excluded book also only blocks re-adding that same title, so an excluded box set no longer makes the volume it is named after impossible to add.
• Series built from Hardcover metadata were never linked to Hardcover (#2245). A series created from a provider series reference got the right hc-series: id and no link row, so it showed as having no Hardcover link, the catalogue diff was unavailable, and Fill silently did nothing. The link is now recorded as the series is created. Series created before this release stay unlinked until you link them from the series Search button.
• A strict unknown-language rule emptied a Hardcover catalogue (#2241). Hardcover's author works query cannot ask for a language, because language belongs to an edition rather than a book, so every work arrived with no language at all. A metadata profile with allowed languages set and unknown-language behaviour set to fail then rejected all of them, and an author with 23 Hardcover series synced to a catalogue of leftovers and no series, with nothing reported as wrong. The language is now derived from the book's default edition. Nothing was deleted by this: the books were never created, so a refresh brings them back. Thanks to kevinatlee for the fix.
• Adding a book could bind it to a cue sheet in another book's folder (#2240). The already-owned check that runs when a book is created never received the supplement ranking added in v1.32.2, so a .cue.txt sitting beside an audiobook could be recorded as the file for a book you do not own, and one cue sheet could be handed to two different books in a single author sync, each with its on-add search silently skipped. That check now applies the same two tier treatment as the library scan: a supplement-class file next to audio is the audiobook's material and never answers, and elsewhere a real container outranks a supplement while still matching when nothing better does, so text-only and PDF-only libraries keep working.
• The provider fallback when adding an author is no longer silent (#2237). With metadata.primary_provider set to Hardcover, adding an author whose record Hardcover's search does not return created an author linked to another provider, and because the catalogue provider is derived from that link the author then synced from the other provider permanently. Nothing said so. The add dialog now flags the result, warns before the add that the catalogue will sync from the other provider, and the API response carries a providerMismatch object.
• Relink endpoints honour tenancy enforcement being off (#2243). relink-upstream and its candidates endpoint returned "author not found" for any caller whose user id differed from the author's owner, even with BINDERY_ENFORCE_TENANCY unset, while every other author endpoint let that same caller read and delete the author. Both now scope their lookup the way their neighbours do. Cross-user access stays blocked when enforcement is on.
• Per-indexer seed ratios reach qBittorrent again (#2205). qBittorrent 5.2 made shareLimitAction a required parameter of the share limits endpoint, so every attempt to apply an indexer's seed ratio was rejected with HTTP 400 and the torrent silently kept the client's global limit. Bindery now sends it, with the value that applies the client's own configured action when a limit is reached, so behaviour is unchanged where it was already working. Older qBittorrent versions ignore the extra field.