diff --git a/CHANGELOG.md b/CHANGELOG.md
index 90db6285..475e7f03 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -9,6 +9,30 @@ generated trace matrix lives in [docs/traceability.md](docs/traceability.md).
 For how long each fixed defect had been shipping before it was found, see
 [docs/defect-windows.md](docs/defect-windows.md).
 
+## v0.11.1
+
+Two faults that had been present since the first release, both found on a
+device: changing the audio track did nothing, and no subtitle would load.
+
+### 🐛 Fixes
+
+- **Changing the audio track changes the audio.** Picking a different language
+  did nothing on Android — the menu closed, the tick moved, and the original
+  track kept playing, with nothing saying otherwise. When the server is
+  converting a film it builds that conversion around *one* audio track, so the
+  others are not in the stream that arrives; the app was asking the player to
+  select from tracks it had never been sent. It now asks the server for the
+  track you picked and resumes where you were. A film playing in its original
+  form still switches instantly, because there every track really is present.
+  (UR-021 → DR-258)
+
+- **Subtitles load.** Every subtitle in the list was inert: the address the app
+  fetched them from was missing a segment, so each request came back "not
+  found", and a subtitle that never arrives is a subtitle the player cannot
+  offer. All of them had been failing this way since the first release — the
+  tests that were supposed to cover the address were checking a copy of it kept
+  inside the tests, not the one being requested. (UR-020 → DR-259)
+
 ## v0.11.0
 
 Video can play through the native renderer on Linux, and the machinery every
diff --git a/docs/defect-windows.md b/docs/defect-windows.md
index 92c8395e..d28075d3 100644
--- a/docs/defect-windows.md
+++ b/docs/defect-windows.md
@@ -17,8 +17,8 @@ row can be re-checked or disputed:
 
 ## Present since the first release
 
-Nine defects date to the initial proof of concept (v0.0.1, 2026-06-23) and shipped
-for between two weeks and seven weeks short of two months before anyone hit them.
+Fifteen defects date to the initial proof of concept (v0.0.1, 2026-06-23) and
+shipped for between two weeks and two months before anyone hit them.
 That is the dominant pattern here: not regressions, but original assumptions that
 went unexercised until a later feature leaned on them.
 
@@ -34,9 +34,12 @@ went unexercised until a later feature leaned on them.
 | No `PlaySessionId`, and one hardcoded `DeviceId`, on every stream URL (DR-177) | v0.0.1 | **v0.5.5** | ~8 weeks | pickaxe |
 | `download_item` never recorded `media_type`; NULL read as `'audio'` (DR-135) | v0.0.1 | **v0.4.6** | ~7 weeks | pickaxe |
 | `download_album` read its track list from the local cache (DR-173) | v0.0.1 | **v0.5.5** | ~8 weeks | pickaxe |
+
 | Device profile carried no `MaxAudioChannels` (DR-141) | v0.0.1 | **v0.4.6** | ~7 weeks | absence |
 | Streaming ceiling fixed at 20 Mbps with no way to lower it (UR-074) | v0.0.1 | **v0.5.3** (as a feature) | ~7.5 weeks | pickaxe |
 | Hero banner auto-rotation never restarted after a manual swipe (DR-038) | v0.0.1 | **v0.9.1** | ~8.5 weeks | pickaxe |
+| Audio-track change asked the player to select a track the transcode never carried (DR-258) | v0.0.1 | **v0.11.1** | ~2 months | pickaxe |
+| Subtitle URL missing its `Stream.` route segment, so every fetch 404ed (DR-259) | v0.0.1 | **v0.11.1** | ~2 months | pickaxe |
 
 ### Why they took so long to surface
 
diff --git a/docs/requirements.md b/docs/requirements.md
index 61c838b3..fea772b3 100644
--- a/docs/requirements.md
+++ b/docs/requirements.md
@@ -451,6 +451,8 @@ Internal architecture, components, and application logic.
 | DR-255 | One helper answers "what URL should an engine open". `playback_url` was gated to Android because only ExoPlayer needed it, and that gate is why a byte-identical copy was later added for the cross-platform open path — the original is invisible in a Linux build, so nothing warned. Two matches over `MediaSource` meant a new variant could be handled in one and forgotten in the other | Player | UR-081 | Done |
 | DR-256 | The video control bar opens **at most one menu at a time**, and opens it where it can be read. Audio track, quality and subtitles each owned a `show…` boolean that no other toggle cleared, so a second menu opened stacked over the first — two panels in the same corner, the newer one covering rows of the older, both still taking clicks. A single `openMenu` value replaces them, which makes "one menu" a property of the state rather than something every handler must remember; the desktop volume popup joins the same group through `VolumeControl`'s optional controlled-open props. Placement was the second half of the same defect: every panel was `absolute right-0` against **its own icon button**, and those icons sit mid-row, so a 220 px panel hung off the left edge of a portrait phone and half the tracks could not be read or tapped. One shared panel now anchors to the control ROW's right edge, clamped to `min(20rem, 100vw − 2rem)` wide and `min(300px, 45vh)` tall, with a full-screen dismiss layer inside the controls subtree so a tap elsewhere closes it without reaching the container's tap gestures (DR-098). The icon row itself wraps instead of overflowing — in portrait the transport controls plus nine icons are wider than the screen, which put fullscreen and close past the edge | UI | UR-020, UR-021, UR-066, UR-074 | Done |
 | DR-257 | A container's children are ordered by **what the container is**, decided in Rust. The frontend pinned `SortBy=SortName` onto every drill-down, so a Jellypod podcast — a Jellyfin channel folder whose plugin returns episodes newest-first and prefixes played ones with "[Played]" — listed alphabetically, which both discarded the release order and clumped every heard episode at the top. `ChannelFolderItem` with `is_folder` now maps to its own `MediaKind::ChannelFolder` rather than collapsing into `Folder`, which is what makes the two distinguishable at all; `default_listing_sort` maps that kind to `PremiereDate` descending and every other container to `SortName` ascending, and a caller that names no container still gets no `SortBy`, so paths relying on the server's own order (a playlist's stored order) keep it. An explicit sort always wins. The offline leg of the cache/server race applies the same order, so the cached list does not flash in name order before the server's arrives. The store now names the container and never a sort field — the ordering rule is domain vocabulary, the same division as `SearchScope` | Repository | UR-007 | Done |
+| DR-258 | An audio-track change is honoured by **re-opening the stream** when the stream cannot carry the track. Jellyfin builds a transcode around one `AudioStreamIndex`, so the alternate tracks are not in it — but the native path only ever called `setAudioTrack(n)`, which indexes ExoPlayer's audio track *groups*. On Android that is the common case, since any source whose default audio codec the device cannot decode is transcoded: ExoPlayer held one audio track while the menu listed every track in the file, so every selection warned `Invalid audio track index` and was dropped, leaving the default track playing with nothing in the UI saying so. `determine_audio_track_switch_strategy` now decides by whether the stream in front of the engine carries the track at all — a direct play still selects in place, a transcode is re-negotiated at the chosen index and resumed. Where it resumes is the player's answer, not the UI's: the native path has no `<video>` element to read, so it sends no position, and defaulting that to zero re-opened the film at the beginning | Player | UR-021, UR-005 | Done |
+| DR-259 | Subtitle URLs address Jellyfin's route, `/Videos/{item}/{source}/Subtitles/{index}/Stream.{format}`. The `Stream.` segment was missing, which matches no route and 404s, so every sideloaded subtitle failed to fetch. Since media3 1.5 a sideloaded text track only becomes a track group once its file is parsed, so 42 failed fetches left ExoPlayer with no text tracks at all and subtitle selection warned `available: 0` and did nothing. The URL tests that existed asserted the shape of a *mock helper* duplicating the format string rather than the URL the app requests, which is why a route error survived from the first release | Repository | UR-020 | Done |
 | DR-198 | The webview runs under a real Content-Security-Policy, and the asset protocol is scoped to the one directory it still serves. `csp` was `null`, which disables CSP entirely: any script that reached the web layer — through a future `{@html}`, a dependency, or a devtools paste — would have inherited the whole IPC surface, and with it the user's session. `script-src 'self'` (Tauri injects a nonce for SvelteKit's inline bootstrap script at build time, so no `'unsafe-inline'` is needed) plus `object-src`/`frame-src 'none'` and `base-uri 'self'` is the part that is genuinely restrictive. `img-src`/`media-src`/`connect-src` cannot be: the Jellyfin origin is typed in by the user at run time and is commonly plain `http` on a LAN, so they allow `http:`/`https:` — a wide grant for *data*, but one that still bars `file:`, `filesystem:` and scripting schemes, and leaves `script-src` untouched. `style-src` keeps `'unsafe-inline'` because Svelte compiles `style="…"` attributes (including `app.html`'s `display: contents` wrapper) into markup; this is safe only while no `<style>` element survives into `index.html`, since a nonce there would make Tauri's injection outrank — and therefore void — `'unsafe-inline'`. `worker-src blob:` and `media-src blob:` are hls.js: it demuxes in a worker built from a blob and attaches MSE through `URL.createObjectURL`. `asset:` and `http://asset.localhost` are the same protocol under the two naming schemes `convertFileSrc` emits (custom scheme on Linux/macOS, `http` host on Windows/Android); `ipc:`/`http://ipc.localhost` is the invoke transport, which would otherwise be blocked by `connect-src`. A run-time CSP naming the server origin exactly was rejected: Tauri computes the header from immutable config when it serves the HTML, so it would mean rebuilding config and reloading the webview on every server change, for a policy the user can already point anywhere. The asset-protocol scope narrows from `$APPDATA/**` to `$APPDATA/thumbnails/**` — since DR-137 moved downloaded media to the loopback server, `imageCache` is the only `convertFileSrc` caller left, so the database and the encrypted-token fallback file no longer sit inside the grant | Security | UR-012, UR-071 | Done |
 
 ---
@@ -774,6 +776,9 @@ Internal architecture, components, and application logic.
 | UT-229 | A channel folder's children are requested by release date, newest first, while every other container keeps name order; an explicit sort still wins, and a caller naming no container gets no `SortBy` at all | DR-257 | Done |
 | UT-230 | A `ChannelFolderItem` that is a folder maps to `ChannelFolder`, not to the generic `Folder` it was indistinguishable from | DR-257 | Done |
 | UT-231 | The library store sends the container's kind and no sort field, defaulting to a plain folder when the caller names none | DR-257 | Done |
+| UT-232 | A transcode's audio-track change re-opens the stream, a direct play selects in place, and an HTML5 element reloads either way — the engine is only asked to select a track the stream actually carries | DR-258 | Done |
+| UT-233 | The position a re-opened stream resumes at comes from the engine when the caller has none, and a non-finite or negative position is treated as absent rather than passed to a backend that rejects it | DR-258 | Done |
+| UT-234 | A subtitle URL targets Jellyfin's `Stream.{format}` route, asserted against the repository that builds it rather than a mock that restates it | DR-259 | Done |
 
 ### Integration Tests
 
diff --git a/src-tauri/src/commands/player/mod.rs b/src-tauri/src/commands/player/mod.rs
index 06e69415..7c9d5f6d 100644
--- a/src-tauri/src/commands/player/mod.rs
+++ b/src-tauri/src/commands/player/mod.rs
@@ -25,8 +25,9 @@ use super::DatabaseWrapper;
 use crate::download::cache::{CacheConfig, SmartCache};
 use crate::jellyfin::{JellyfinClient, JellyfinConfig};
 use crate::player::{
-    determine_video_seek_strategy, MediaItem, MediaSessionManager, MediaSource, MediaType,
-    PlayerController, PlayerState, PlayerStatusEvent, QueueContext, RepeatMode, VideoSeekStrategy,
+    determine_audio_track_switch_strategy, determine_video_seek_strategy, AudioTrackSwitchStrategy,
+    MediaItem, MediaSessionManager, MediaSource, MediaType, PlayerController, PlayerState,
+    PlayerStatusEvent, QueueContext, RepeatMode, VideoSeekStrategy,
 };
 use crate::repository::{
     types::{GetItemsOptions, ImageOptions, ImageType},
@@ -1595,18 +1596,37 @@ pub async fn player_seek_video(
     }
 }
 
-/// Switch audio track - handles both HTML5 (stream reload) and native (direct switch)
+/// Switch audio track.
 /// Note: Frontend should handle saving series preferences after this command succeeds
 ///
-/// The split is the requirement: an HTML5 `<video>` element cannot be told to
-/// change audio track, so the stream is re-opened at the chosen
-/// `AudioStreamIndex` and the frontend seeks the reloaded element back to
-/// `position`; a native backend (ExoPlayer) switches in place by track-group
-/// index. libmpv implements neither — it is the audio-only backend here and
-/// leaves `PlayerBackend::set_audio_track` at its `not_implemented()` default,
-/// which is why IR-019 is met by these two paths rather than by MPV.
+/// What decides the route is **whether the stream in front of the engine
+/// carries the requested track at all** — see
+/// [`determine_audio_track_switch_strategy`]:
 ///
-/// TRACES: UR-021 | IR-019, DR-024
+/// - An HTML5 `<video>` element has no track-selection API, so the stream is
+///   always re-opened at the chosen `AudioStreamIndex` and the frontend seeks
+///   the reloaded element back to `position`.
+/// - A native backend playing a **direct play** holds the source file with
+///   every track in it, so ExoPlayer selects in place by track-group index.
+/// - A native backend playing a **transcode** does not. Jellyfin builds a
+///   transcode around one `AudioStreamIndex`, so the alternate tracks are not
+///   in the stream; the switch has to re-open it, which this command does
+///   itself and resumes at `current_position`.
+///
+/// That last case is a bug fix, and it was the common case on Android: any
+/// source whose default audio codec the device cannot decode is transcoded, so
+/// ExoPlayer saw `Audio tracks: 1` while the menu listed every track in the
+/// file. The old code called `setAudioTrack(n)` regardless, which indexes
+/// ExoPlayer's audio track *groups*, found nothing at `n`, warned `Invalid
+/// audio track index` and dropped the request — the default track just kept
+/// playing, with nothing in the UI saying so.
+///
+/// libmpv implements neither selection nor reload here — it is the audio-only
+/// backend and leaves `PlayerBackend::set_audio_track` at its
+/// `not_implemented()` default, which is why IR-019 is met by these paths
+/// rather than by MPV.
+///
+/// TRACES: UR-021, UR-005 | IR-019, DR-024, DR-258
 #[tauri::command]
 #[specta::specta]
 // Two of the eight arguments are Tauri `State<'_, _>` injections, not caller
@@ -1626,56 +1646,128 @@ pub async fn player_switch_audio_track(
     info!("[player_switch_audio_track] Switching to audio track - stream_index: {}, array_index: {}, use_html5: {}",
         stream_index, array_index, use_html5);
 
-    if use_html5 {
-        // HTML5 backend needs stream reload
-        let repository = repository_manager
-            .0
-            .get(&repository_handle)
-            .ok_or("Repository not found - user may need to log in")?;
+    // Read what the engine is playing before deciding anything — including
+    // where it is, which has to be captured before the stop below wipes it.
+    // Locks are dropped at the end of this block so none is held across an
+    // await.
+    let (jellyfin_item_id, needs_transcoding, engine_position) = {
+        let controller = player.0.lock().await;
+        let engine_position = controller.absolute_position();
+        let queue_arc = controller.queue();
+        let queue = queue_arc.lock().map_err(|e| e.to_string())?;
 
-        // Get current item to find Jellyfin ID
-        let jellyfin_item_id = {
-            let controller = player.0.lock().await;
-            let queue_arc = controller.queue();
-            let queue = queue_arc.lock().map_err(|e| e.to_string())?;
-
-            let current_item = queue.current().ok_or("No item currently playing")?;
+        let current_item = queue.current().ok_or("No item currently playing")?;
 
+        (
             current_item
                 .jellyfin_id()
                 .ok_or("Current item has no Jellyfin ID")?
-                .to_string()
-        };
+                .to_string(),
+            current_item.needs_transcoding,
+            engine_position,
+        )
+    };
 
-        // Select a stream carrying the chosen audio track. It starts at zero —
-        // an HLS playlist cannot carry a position (DR-181) — and `position`
-        // below tells the frontend where to seek the reloaded element back to.
-        //
-        // Pinning a track is itself a reason the source cannot be direct-played:
-        // the file has one default track and the viewer asked for another, so
-        // the negotiation returns a transcode. That decision lives in
-        // `decide_playback_kind`, not here.
-        let selection = repository
-            .get_stream_selection(
-                &jellyfin_item_id,
-                media_source_id.as_deref(),
-                Some(stream_index),
-            )
-            .await
-            .map_err(|e| format!("Failed to select a stream: {:?}", e))?;
+    let strategy = determine_audio_track_switch_strategy(needs_transcoding, use_html5);
 
-        Ok(AudioTrackSwitchResponse::ReloadStream {
-            selection,
-            position: current_position.unwrap_or(0.0),
-        })
-    } else {
-        // Native backend (Android ExoPlayer) - use array index
+    info!(
+        "[player_switch_audio_track] needs_transcoding={}, use_html5={}, strategy={:?}",
+        needs_transcoding, use_html5, strategy
+    );
+
+    if strategy == AudioTrackSwitchStrategy::BackendSelectInPlace {
+        // A direct play: the engine holds the source file, every track included.
         let controller = player.0.lock().await;
         controller
             .set_audio_track(array_index)
             .map_err(|e| e.to_string())?;
 
-        Ok(AudioTrackSwitchResponse::Native { success: true })
+        return Ok(AudioTrackSwitchResponse::Native { success: true });
+    }
+
+    // Both reload strategies need a stream built around the chosen track.
+    let repository = repository_manager
+        .0
+        .get(&repository_handle)
+        .ok_or("Repository not found - user may need to log in")?;
+
+    // Select a stream carrying the chosen audio track. It starts at zero —
+    // an HLS playlist cannot carry a position (DR-181) — so the position is
+    // restored by seeking afterwards, here or in the frontend.
+    //
+    // Pinning a track is itself a reason the source cannot be direct-played:
+    // the file has one default track and the viewer asked for another, so
+    // the negotiation returns a transcode. That decision lives in
+    // `decide_playback_kind`, not here.
+    let selection = repository
+        .get_stream_selection(
+            &jellyfin_item_id,
+            media_source_id.as_deref(),
+            Some(stream_index),
+        )
+        .await
+        .map_err(|e| format!("Failed to select a stream: {:?}", e))?;
+
+    // The caller's position if it has one, the engine's otherwise. The native
+    // path has no `<video>` element to read, so it sends none — and defaulting
+    // that to zero re-opened the stream at the start of the film.
+    let position = crate::player::track_switch::resume_position(current_position, engine_position);
+
+    match strategy {
+        AudioTrackSwitchStrategy::Html5ReloadStream => Ok(AudioTrackSwitchResponse::ReloadStream {
+            selection,
+            position,
+        }),
+        AudioTrackSwitchStrategy::BackendReloadStream => {
+            // The native backend re-opens its own stream, the same sequence the
+            // transcoded seek and quality change use: stop, repoint the queue
+            // entry at the new URL, load, then seek back to where the viewer
+            // was. Nothing is left for the frontend to do.
+            let new_url = selection.url.clone();
+
+            {
+                let controller = player.0.lock().await;
+                controller.stop().map_err(|e| e.to_string())?;
+            }
+
+            {
+                let controller = player.0.lock().await;
+                let queue_arc = controller.queue();
+                let mut queue = queue_arc.lock().map_err(|e| e.to_string())?;
+
+                if !queue.update_current_stream_url(new_url) {
+                    return Err("Failed to update stream URL in queue".to_string());
+                }
+            }
+
+            {
+                let controller = player.0.lock().await;
+                let queue_arc = controller.queue();
+                let queue = queue_arc.lock().map_err(|e| e.to_string())?;
+
+                let updated_item = queue
+                    .current()
+                    .ok_or("No current item after URL update")?
+                    .clone();
+                drop(queue);
+
+                controller
+                    .load_and_play(&updated_item)
+                    .map_err(|e| e.to_string())?;
+                controller.seek(position).map_err(|e| e.to_string())?;
+            }
+
+            info!(
+                "[player_switch_audio_track] Re-opened the stream on audio stream {} and resumed at {}",
+                stream_index, position
+            );
+
+            Ok(AudioTrackSwitchResponse::Native { success: true })
+        }
+        // Handled above, before the stream was negotiated.
+        AudioTrackSwitchStrategy::BackendSelectInPlace => {
+            Ok(AudioTrackSwitchResponse::Native { success: true })
+        }
     }
 }
 
diff --git a/src-tauri/src/player/mod.rs b/src-tauri/src/player/mod.rs
index 414c6445..30897d90 100644
--- a/src-tauri/src/player/mod.rs
+++ b/src-tauri/src/player/mod.rs
@@ -23,6 +23,7 @@ pub mod session;
 pub mod sleep_timer;
 pub mod state;
 pub mod stream_end;
+pub mod track_switch;
 
 #[cfg(test)]
 mod mpv_backend_test;
@@ -71,6 +72,7 @@ pub use seek::{determine_video_seek_strategy, VideoSeekStrategy};
 pub use session::{MediaSessionManager, MediaSessionType};
 pub use sleep_timer::{SleepTimerMode, SleepTimerState};
 pub use state::{EndReason, PlayerState};
+pub use track_switch::{determine_audio_track_switch_strategy, AudioTrackSwitchStrategy};
 
 // Re-export platform-specific backends
 #[cfg(target_os = "android")]
diff --git a/src-tauri/src/player/track_switch.rs b/src-tauri/src/player/track_switch.rs
new file mode 100644
index 00000000..5f77ad61
--- /dev/null
+++ b/src-tauri/src/player/track_switch.rs
@@ -0,0 +1,155 @@
+//! Audio-track switch strategy decision logic.
+//!
+//! Pure logic, extracted from the command layer so it can be unit-tested in the
+//! player core — the sibling of [`super::seek`]. `player_switch_audio_track`
+//! turns the resulting [`AudioTrackSwitchStrategy`] into a concrete action.
+//!
+//! The rule this module exists to state: **an engine can only select a track
+//! the stream in front of it actually carries.** A Jellyfin transcode is built
+//! around one `AudioStreamIndex`, so the alternate tracks are not in the stream
+//! at all — the switch has to re-open it. Only a direct play/stream hands the
+//! engine the source file with every track present.
+
+/// How a request to change audio track has to be carried out.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum AudioTrackSwitchStrategy {
+    /// Re-open the stream pinned to the chosen track; the frontend reloads its
+    /// `<video>` element. An HTML5 element cannot select an audio track at all,
+    /// so this holds whether or not the current stream is a transcode.
+    Html5ReloadStream,
+    /// Re-open the stream pinned to the chosen track; the backend reloads
+    /// itself and restores the position.
+    BackendReloadStream,
+    /// The engine already holds every track — select in place, no reload.
+    BackendSelectInPlace,
+}
+
+/// Decide how to honour an audio-track change.
+///
+/// # Arguments
+/// * `needs_transcoding` - Whether the stream now playing is a server-side
+///   transcode, which carries exactly the one audio track it was built around.
+/// * `use_html5` - Whether the frontend `<video>` element is rendering.
+///
+/// TRACES: UR-021 | IR-019, DR-024, DR-258 | UT-232
+pub fn determine_audio_track_switch_strategy(
+    needs_transcoding: bool,
+    use_html5: bool,
+) -> AudioTrackSwitchStrategy {
+    if use_html5 {
+        return AudioTrackSwitchStrategy::Html5ReloadStream;
+    }
+
+    if needs_transcoding {
+        AudioTrackSwitchStrategy::BackendReloadStream
+    } else {
+        AudioTrackSwitchStrategy::BackendSelectInPlace
+    }
+}
+
+/// Where to resume after re-opening the stream for a track change.
+///
+/// `requested` is what the caller supplied; `engine_position` is where the
+/// engine itself says it is. The caller wins when it has something real to say,
+/// and the engine answers otherwise — which is the whole point: **position is
+/// the player's to know**, not the UI's to remember.
+///
+/// The native path proved why. It has no `<video>` element, so the frontend
+/// sent `null`, the command defaulted to `0.0`, and switching audio track
+/// re-opened the stream at the beginning of the film — the track changed and
+/// the viewer lost their place. A non-finite or negative value is treated the
+/// same as absent rather than passed through to a backend that would reject it.
+///
+/// TRACES: UR-021, UR-005 | IR-019, DR-024, DR-258 | UT-233
+pub fn resume_position(requested: Option<f64>, engine_position: f64) -> f64 {
+    let usable = requested.filter(|p| p.is_finite() && *p > 0.0);
+    let fallback = if engine_position.is_finite() && engine_position > 0.0 {
+        engine_position
+    } else {
+        0.0
+    };
+
+    usable.unwrap_or(fallback)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// The reported bug, seen on a device: switching audio track changed the
+    /// track but "restarts from zero". The native path has no `<video>`
+    /// element, so the frontend passed `null` and the re-opened stream began at
+    /// the start of the film — logcat: `Re-opened the stream on audio stream 2
+    /// and resumed at 0` while playback was 22 minutes in.
+    #[test]
+    fn a_caller_with_no_position_resumes_where_the_engine_is() {
+        assert_eq!(resume_position(None, 1337.5), 1337.5);
+    }
+
+    /// The HTML5 path does have an element and its clock is the honest answer
+    /// there, so what the caller supplies wins.
+    #[test]
+    fn a_caller_that_knows_its_position_is_believed() {
+        assert_eq!(resume_position(Some(42.0), 1337.5), 42.0);
+    }
+
+    /// A position that is not a position — NaN from an element with no
+    /// metadata, or a negative from a clock read mid-teardown — is treated as
+    /// absent. Passing it through re-opens at a place no backend accepts.
+    #[test]
+    fn a_nonsense_position_falls_back_to_the_engine() {
+        assert_eq!(resume_position(Some(f64::NAN), 90.0), 90.0);
+        assert_eq!(resume_position(Some(-5.0), 90.0), 90.0);
+        assert_eq!(resume_position(None, f64::NAN), 0.0);
+    }
+
+    /// Switching track in the first moments of playback resumes at the start,
+    /// which is where the viewer actually is.
+    #[test]
+    fn the_very_beginning_stays_the_very_beginning() {
+        assert_eq!(resume_position(None, 0.0), 0.0);
+    }
+
+    /// The reported bug: on Android the audio-track menu did nothing and the
+    /// default track kept playing.
+    ///
+    /// Jellyfin had negotiated a transcode (`TranscodeReasons=AudioCodecNot
+    /// Supported`) whose URL pins `AudioStreamIndex=1`, so ExoPlayer was handed
+    /// a stream with exactly one audio track — logcat: `Audio tracks: 1`. The
+    /// native path nonetheless only ever called `setAudioTrack(n)`, which
+    /// indexes ExoPlayer's audio track *groups* and so found nothing to select:
+    /// `Invalid audio track index: 1 (available: 1)`, warned and dropped. The
+    /// track the viewer asked for is not in the stream; it has to be re-opened.
+    #[test]
+    fn a_transcode_is_re_opened_because_it_carries_only_one_track() {
+        assert_eq!(
+            determine_audio_track_switch_strategy(true, false),
+            AudioTrackSwitchStrategy::BackendReloadStream
+        );
+    }
+
+    /// A direct play hands the engine the source file, every track included, so
+    /// ExoPlayer selects in place — no reload, no re-buffer, no lost position.
+    #[test]
+    fn a_direct_play_switches_in_place() {
+        assert_eq!(
+            determine_audio_track_switch_strategy(false, false),
+            AudioTrackSwitchStrategy::BackendSelectInPlace
+        );
+    }
+
+    /// An HTML5 `<video>` element has no track-selection API, so it reloads
+    /// either way. This is the path that already worked, and it must keep
+    /// working: the fix is about the native side only.
+    #[test]
+    fn html5_always_reloads_because_the_element_cannot_select() {
+        assert_eq!(
+            determine_audio_track_switch_strategy(true, true),
+            AudioTrackSwitchStrategy::Html5ReloadStream
+        );
+        assert_eq!(
+            determine_audio_track_switch_strategy(false, true),
+            AudioTrackSwitchStrategy::Html5ReloadStream
+        );
+    }
+}
diff --git a/src-tauri/src/repository/online.rs b/src-tauri/src/repository/online.rs
index 2858e539..d628a0e4 100644
--- a/src-tauri/src/repository/online.rs
+++ b/src-tauri/src/repository/online.rs
@@ -2469,8 +2469,15 @@ impl MediaRepository for OnlineRepository {
         stream_index: i32,
         format: &str,
     ) -> String {
+        // `Stream.{format}` is the route, not a filename we get to choose:
+        // Jellyfin exposes the subtitle as
+        // `/Videos/{item}/{source}/Subtitles/{index}/Stream.{format}`, and
+        // stopping at the format alone matches no route and 404s. Every
+        // sideloaded subtitle failed to load on Android because of it, leaving
+        // ExoPlayer with no text tracks to select.
+        // TRACES: UR-020 | JA-008, DR-259 | UT-234
         format!(
-            "{}/Videos/{}/{}/Subtitles/{}/{}",
+            "{}/Videos/{}/{}/Subtitles/{}/Stream.{}",
             self.server_url, item_id, media_source_id, stream_index, format
         )
     }
@@ -3019,6 +3026,35 @@ mod tests {
         )
     }
 
+    /// The reported bug: on Android every subtitle track was inert — the menu
+    /// listed 42 languages and picking one changed nothing.
+    ///
+    /// The cause is here rather than in the player. ExoPlayer sideloads each
+    /// subtitle as its own media source, and since media3 1.5 a sideloaded text
+    /// track only becomes a *track group* once its file has been fetched and
+    /// parsed. Every fetch 404ed, so `Tracks` carried no text group at all and
+    /// `setSubtitleTrack(1)` warned `available: 0` and dropped the request.
+    ///
+    /// Jellyfin's route is `/Videos/{item}/{source}/Subtitles/{index}/Stream.{fmt}`
+    /// (verified against a live server: this shape answers 200, the one built
+    /// here answered 404). The `Stream.` segment is not decoration — without it
+    /// the path matches no route.
+    ///
+    /// The old mock-based URL tests could not catch this: they asserted the
+    /// shape of a *test helper* that duplicated the format string, not of the
+    /// URL the app actually requests.
+    ///
+    /// TRACES: UR-020 | JA-008, DR-259 | UT-234
+    #[test]
+    fn subtitle_url_uses_jellyfins_stream_route() {
+        let repo = create_test_repository();
+
+        assert_eq!(
+            repo.get_subtitle_url("item123", "source456", 2, "vtt"),
+            "https://test.server.com/Videos/item123/source456/Subtitles/2/Stream.vtt"
+        );
+    }
+
     /// Build a repository wired to a real ConnectivityReporter so we can assert
     /// how `report_outcome` classifies each `RepoError` into reachability.
     /// (No app handle → event emission is a harmless no-op.)
diff --git a/src/lib/api/bindings.ts b/src/lib/api/bindings.ts
index 3c83f716..e5dcbe19 100644
--- a/src/lib/api/bindings.ts
+++ b/src/lib/api/bindings.ts
@@ -145,18 +145,37 @@ async playerSetAudioTrack(streamIndex: number) : Promise<PlayerStatus> {
     return await TAURI_INVOKE("player_set_audio_track", { streamIndex });
 },
 /**
- * Switch audio track - handles both HTML5 (stream reload) and native (direct switch)
+ * Switch audio track.
  * Note: Frontend should handle saving series preferences after this command succeeds
  * 
- * The split is the requirement: an HTML5 `<video>` element cannot be told to
- * change audio track, so the stream is re-opened at the chosen
- * `AudioStreamIndex` and the frontend seeks the reloaded element back to
- * `position`; a native backend (ExoPlayer) switches in place by track-group
- * index. libmpv implements neither — it is the audio-only backend here and
- * leaves `PlayerBackend::set_audio_track` at its `not_implemented()` default,
- * which is why IR-019 is met by these two paths rather than by MPV.
+ * What decides the route is **whether the stream in front of the engine
+ * carries the requested track at all** — see
+ * [`determine_audio_track_switch_strategy`]:
  * 
- * TRACES: UR-021 | IR-019, DR-024
+ * - An HTML5 `<video>` element has no track-selection API, so the stream is
+ * always re-opened at the chosen `AudioStreamIndex` and the frontend seeks
+ * the reloaded element back to `position`.
+ * - A native backend playing a **direct play** holds the source file with
+ * every track in it, so ExoPlayer selects in place by track-group index.
+ * - A native backend playing a **transcode** does not. Jellyfin builds a
+ * transcode around one `AudioStreamIndex`, so the alternate tracks are not
+ * in the stream; the switch has to re-open it, which this command does
+ * itself and resumes at `current_position`.
+ * 
+ * That last case is a bug fix, and it was the common case on Android: any
+ * source whose default audio codec the device cannot decode is transcoded, so
+ * ExoPlayer saw `Audio tracks: 1` while the menu listed every track in the
+ * file. The old code called `setAudioTrack(n)` regardless, which indexes
+ * ExoPlayer's audio track *groups*, found nothing at `n`, warned `Invalid
+ * audio track index` and dropped the request — the default track just kept
+ * playing, with nothing in the UI saying so.
+ * 
+ * libmpv implements neither selection nor reload here — it is the audio-only
+ * backend and leaves `PlayerBackend::set_audio_track` at its
+ * `not_implemented()` default, which is why IR-019 is met by these paths
+ * rather than by MPV.
+ * 
+ * TRACES: UR-021, UR-005 | IR-019, DR-024, DR-258
  */
 async playerSwitchAudioTrack(repositoryHandle: string, streamIndex: number, arrayIndex: number, useHtml5: boolean, currentPosition: number | null, mediaSourceId: string | null) : Promise<AudioTrackSwitchResponse> {
     return await TAURI_INVOKE("player_switch_audio_track", { repositoryHandle, streamIndex, arrayIndex, useHtml5, currentPosition, mediaSourceId });