diff --git a/CHANGELOG.md b/CHANGELOG.md
index 141c104f5..c797ae58e 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -16,7 +16,7 @@ All notable changes to Bambuddy will be documented in this file.
 - **Git backup now actually writes cloud profiles (#2717, reporter @jmoore-skild)** — Enabling **Cloud Profiles** for a Git backup produced nothing. The collector looked for a `setting` list in the Bambu Cloud response, which is keyed by preset type instead, so the loop never ran once — and it asked for the credential store used when authentication is *disabled*, so on any install with authentication on it found no account to collect from in the first place. Neither failure was visible: `backup_metadata.json` still recorded `cloud_profiles: true`, and the log line read `Collected cloud profiles: 0 filament, 0 printer, 0 process`, which looks like a successful backup of an empty account. Cloud profiles are now collected from **every connected account across both Bambu Cloud and Orca Cloud**, one directory per cloud per account, keyed by user ID so no email address is written into a backup repository. Bambu presets are stored with the payload needed to recreate them rather than just their names, and Bambu's bundled public catalogue is skipped — it is identical for everyone, re-downloadable, and would rewrite the repository on every run. The metadata now records what was actually collected, per cloud and per account, and a run that collects nothing while the category is enabled says so as a warning instead of an INFO line that reads like success. The Cloud Profiles checkbox no longer keys off your own Bambu sign-in — it enables when *any* account is connected and shows how many are in scope, which matters on a multi-user install where the presets being backed up are other people's. A backup also no longer disconnects an Orca Cloud account whose session it can't refresh: Orca reports every rejection with one composite reason, so a genuine revocation is indistinguishable from a lost token-rotation race, and an unattended job should not be the thing that guesses. The account is skipped with a warning, and the dead credentials are cleared the next time you open Orca Cloud Profiles — where you can pair again on the spot. Translated in all locales; wiki updated. Covered by backend tests.
 
 ### Added
-- **One queue item, several printer models — whichever frees up first (#671, reporter @brainomite; also delivers most of #2570, reporter @NeighborGeek)** — With an H2S and an H2C, a job you don't care which machine runs still had to be queued twice: the two printers need different slices, a queue item held exactly one file, and "any H2S" and "any H2C" were separate jobs competing for the same plastic. Whichever started first, you deleted the other by hand. Select both sliced files in the File Manager and press **Print** and you now get **one** queue item carrying both — the scheduler walks them in the order you arranged and takes the first whose model has an idle printer. The many-to-many never leaves the scheduler's selection loop: the moment a candidate wins, its file, plate and nozzle mapping are folded onto the queue row, so the upload, archive creation, print history and reprint all see an ordinary single-file job and behave exactly as they always have. Order is yours to set, because "both are free right now" has to resolve the same way every time rather than following whichever match the matcher happened to see first. Candidates are otherwise tried least-attempted first, so a printer that accepts the file and never starts hands the job to the other machine on the next lap instead of spending the item's whole retry budget on the one that is wedged. The set is validated as a set: one file per printer model (two slices for the same machine are not alternatives, and picking between them arbitrarily would look like a bug the first time it chose your draft profile), every file gated against the model it is offered as, and at least one model that actually has a printer — grouping the H2C slice before the H2C arrives is fine, queueing a job nothing can ever run is not. A cross-model item deliberately holds no file of its own, so deleting one alternative leaves the job and its sibling intact; deleting or trashing every candidate holds it with an explanation instead of failing deep in the upload. Files can also be grouped permanently with **Group as versions**, after which printing any one of them offers the others without re-selecting — this is the grouping and the print-time file matching asked for in #2570, minus its nested File Manager listing. An existing library arrives with its groups already built, from slice provenance Bambuddy has been recording since the Slice button shipped and had never read back. Translated in all locales; wiki updated. Covered by backend and frontend tests.
+- **One queue item, several printer models — whichever frees up first (#671, reporter @brainomite; also delivers most of #2570, reporter @NeighborGeek)** — With an H2S and an H2C, a job you don't care which machine runs still had to be queued twice: the two printers need different slices, a queue item held exactly one file, and "any H2S" and "any H2C" were separate jobs competing for the same plastic. Whichever started first, you deleted the other by hand. Select both sliced files in the File Manager and press **Print** and you now get **one** queue item carrying both — the scheduler walks them in the order you arranged and takes the first whose model has an idle printer. The many-to-many never leaves the scheduler's selection loop: the moment a candidate wins, its file, plate and nozzle mapping are folded onto the queue row, so the upload, archive creation, print history and reprint all see an ordinary single-file job and behave exactly as they always have. Order is yours to set, because "both are free right now" has to resolve the same way every time rather than following whichever match the matcher happened to see first. Candidates are otherwise tried least-attempted first, so a printer that accepts the file and never starts hands the job to the other machine on the next lap instead of spending the item's whole retry budget on the one that is wedged. The set is validated as a set: one file per printer model (two slices for the same machine are not alternatives, and picking between them arbitrarily would look like a bug the first time it chose your draft profile), every file gated against the model it is offered as, and at least one model that actually has a printer — grouping the H2C slice before the H2C arrives is fine, queueing a job nothing can ever run is not. A cross-model item deliberately holds no file of its own, so deleting one alternative leaves the job and its sibling intact; deleting or trashing every candidate holds it with an explanation instead of failing deep in the upload. Filament overrides offer everything loaded across **all** the candidate models rather than just the first — a spool loaded on only one of them is still a legitimate choice, it simply narrows which candidates can match — while AMS slot mapping is absent exactly as it is on an ordinary "Any [model]" job, because no printer has been picked yet and the scheduler derives the mapping against whichever one it takes. In the queue the job reads **Any H2D / X1C**, naming every model it is waiting on rather than filing itself under one it may never run on, and its waiting reason is given per model (`H2D: Busy: H2D-1; X1C: No matching material/color`), collapsing to a plain busy message — and no notification — when every model is merely printing. The alternatives are fixed once queued: the schedule, quantity and print options stay editable, but assigning a specific printer or narrowing to one model is refused by both the dialog and the API, since an item holding alternatives *and* a printer would dispatch down the fixed-printer path with no file to send. Cancel and re-queue to change the set. Files can also be grouped permanently with **Group as versions**, after which printing any one of them offers the others without re-selecting — this is the grouping and the print-time file matching asked for in #2570, minus its nested File Manager listing. An existing library arrives with its groups already built, from slice provenance Bambuddy has been recording since the Slice button shipped and had never read back. Translated in all locales; wiki updated. Covered by backend and frontend tests.
 - **Uploaded archives can be named after the filename you sent (#2610, contributor @Person2099)** — A 3MF carries a `print_name` inside its metadata, and that is the name an archive gets. It is often the wrong one: whoever originally sliced the file baked their own title into it, and that title survives every rename afterwards, so a file you deliberately named after the job it belongs to still lands in the archive list under a stranger's label. Bambuddy could already prefer the filename — the FTP review flow and virtual-printer dispatch both do, driven by the virtual printer's **Archive name source** setting — but the upload endpoints could not, so anything sent through the API was stuck with the embedded title. `POST /archives/upload` and `POST /archives/upload-bulk` now take an optional `prefer_filename_for_name` query parameter that names the archive after the uploaded file instead; on the bulk route it applies to every file in the batch. It is off by default, so nothing changes for existing callers or for the Web UI's own upload, which does not set it. A per-request parameter rather than a setting, because the caller — typically an integration naming files after its own jobs — is the only party that knows whether the filename it sent is the meaningful one. Both parameters are documented in the OpenAPI schema, so they show up in `/docs`. Covered by backend tests.
 - **Keep the AMS slots the slicer picked (#2700, contributor @Striker72rus)** — Bambu Studio and OrcaSlicer resolve which physical AMS tray feeds each filament themselves, right before sending. Bambuddy threw that away: a queue-mode virtual printer worked the mapping out again at dispatch time, from the filament type and colour baked into the 3MF. That is usually the better answer — it is computed against the printer's live trays and it respects **Prefer lowest filament** and the AMS-backup gate that goes with it — but it has nothing to go on when the match isn't unique. Two spools of the same red PLA, and the slot you deliberately chose in the slicer is a coin toss. A new per-virtual-printer **Save AMS mapping** toggle keeps the slicer's pick instead: the print dispatches to exactly those trays, and the mapping is stored on the archive so a reprint can reuse the same physical spools — a **Mapping** button in the print modal selects every slot from it in one click, and the archive card and queue row say so. Because a tray number only means something on the AMS it was resolved against, the mapping records its printer and is only ever offered on that same printer; a model-based ("Any [model]") virtual printer has no fixed printer and is unaffected. Off by default, so nothing changes for existing virtual printers until you turn it on, and **Force color match** still wins for the print being dispatched when both are on. Translated in all locales; wiki updated. Covered by backend and frontend tests.
 - **P2S/X2D accessory fans: left auxiliary cooling and chamber exhaust (#2691, contributor @gzimbric, requested in #2660)** — The P2S and X2D have two fans Bambuddy could not show or drive. The **left auxiliary part cooling fan** had no tile and no control at all, because the printer only reports it inside its air-duct data and never in the ordinary fan fields Bambuddy was reading. The **chamber exhaust fan** had the opposite problem: its tile appeared on every P2S whether or not the fan was fitted, so owners of a base machine had a control that did nothing. Both are add-on kits on the P2S and fitted at the factory on the X2D. Both tiles now appear only when the printer itself reports the hardware, so a base P2S looks exactly as it does today and a kitted one gains the fans it actually has. The left auxiliary fan is set from the same speed popover as the others, and the enclosure fan is labelled **Exhaust** on the P2S and X2D — matching the printer's own screen and Bambu Studio — while every other enclosed model keeps **Chamber Fan**. The confirmation message after changing a speed uses the same name as the tile that was clicked. The four tiles are ordered part cooling, left auxiliary, auxiliary, exhaust, so they read left to right in the same order as the physical fans. Both fields are also published through the status endpoint, the WebSocket feed and the MQTT relay, so external automations can read them. Translated in all locales; wiki updated. Covered by backend and frontend tests, including the case where the printer sends a partial fan report — a tile must not disappear mid-print just because one update didn't mention it.