The recommendation metrics table was observable but not actionable: raw
source strings (album:<uuid> one-offs) drowned the stable surfaces, and
manual plays were excluded so skip rates had no control group.
- SQL: include NULL-source rows (the baseline) and carry completion_n
so family merges can weight avg_completion correctly.
- Handler buckets raw sources into stable families (radio:<uuid> →
Radio, album:/artist: → direct plays, etc.) grouped by surface
intent: go-to / discovery / direct — each band judged against its
job, since discovery mixes are expected to skip hotter. Families
under 20 plays are flagged low-confidence, not hidden.
- Settings card renders the baseline row and per-surface deltas in
percentage points vs baseline (worse-than-baseline deltas in danger
color), intent hint copy per group, low-data rows dimmed.
- Pure-unit test for the bucketing/merge; DB test updated to the new
contract (baseline included, radio:<uuid> collapse).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TsF3cNoKrqCYsU78cXC8U6
Confirmed against prod: exactly one track (17 plays, cold since May 21)
met the c>=5 + 30d-cold bar, and three process defects turned that into
a 1-track playlist instead of the locked placeholder.
- ListRediscoverTracks: collapse the two-tier UNION into one blended
pool. The old shallow-tier gate (WHERE NOT EXISTS deep) was
all-or-nothing — one 6-month row suppressed the entire 30-day tier —
and deep was a strict subset of shallow anyway. Eligibility drops to
>=3 non-skip plays (on a weeks-old history the >=5-play tracks are
precisely the ones still in rotation); ordering prefers >=6mo cold,
then >=5 plays, then raw count.
- Minimum viable mix floor for all five discovery mixes: below
minLen (15; 5 for the album-coherent NewForYou/FirstListens) the
variant is withheld so Home renders the 'listen more to unlock'
placeholder instead of a mix that reads as built-wrong.
- /api/events: clamp client-supplied 'at' to [user.created_at,
now+5m]. Unbounded client clocks could write arbitrarily old plays
and poison the 6-month ordering (prod data verified clean — no
scrub needed).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TsF3cNoKrqCYsU78cXC8U6
Failed loads used to stay failed forever: no screen ViewModel listened
to server-health recovery, and the cache-first screens swallowed refresh
errors so a cold load against a down server looked like an empty account.
- connectivity/Recovery.kt: recoveries() — per-collector flow of
down→Healthy transitions; the screen-level half of the idiom
SyncController/MutationReplayer/DiagnosticsUploader already use.
- Every screen VM now re-runs its load on recovery (cache-first screens
unconditionally; direct-load screens when sitting in Error).
- Cache-first error surfacing: Home / Playlists / Liked track refresh
failure; Library reads SyncController.lastSyncError (new) — empty
cache + failed refresh now renders Error-with-Retry, not welcome copy.
- Requests: 12s poll also retries from Error (was structurally unable
to escape it — the in-flight predicate required a Success state).
- Search: retry() bypasses the distinctUntilChanged query pipeline so a
same-text resubmit after a transient failure actually re-runs.
- ArtistDetail: secondary sections (similar artists / top tracks)
re-fetch on recovery instead of staying silently absent.
- ErrorRetry: LazyColumn wrapper (pull-to-refresh works on error states,
same rationale as EmptyState) + optional title; adopted on every error
branch; PlaylistDetail's one-shot ErrorBlock removed in its favor.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TsF3cNoKrqCYsU78cXC8U6
Numeric indices wobble across re-casts (offset +1↔0 seen during output
toggling), making "same track?" ambiguous. Enrich both the track_change
event and the heartbeat with local_track_id (TrackRef.id) and sonos_uri
(RemotePlayerState.currentTrackUri — the URL the speaker is actually
streaming), so a desync is unambiguous.
Also fixes the cast→phone stale-state pollution (#1211): sonos_* is now
zeroed unless a remote route is active, via a shared putSonos() helper —
so a just-ended cast's RemotePlayerState can't masquerade as live Sonos
data in the diagnostics.
Refs Scribe M9 (#119), tasks #1210#1211.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
Heartbeats are 45s apart and missed a rapid skip burst (local_index
16→22 in one gap). Add a 'playback' track_change event emitted on each
queue-index / current-track change, snapshotting local vs Sonos
index+position + server_health + upnp_loading + route — so a transient
skip-induced desync is captured at the instant it happens. (uiState is a
conflated StateFlow, so a very rapid burst may coalesce intermediate
indices; we still get the boundaries + the snapshot.)
Refs Scribe M9 (#119), task #1210.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
Relabel (#1204): route + player_state events fire for every output route,
not just UPnP — split them into a new 'playback' kind; 'upnp_sync' now
means genuinely UPnP/Sonos signal (drops, resync). Migration 0037 adds
'playback' to the kind CHECK; server whitelist, Android reporter labels,
and the web kind filter updated.
Web sort: the diagnostics list gains a Newest/Oldest-first sort (default
newest at top); export follows the displayed order.
Fix (#1205): OutputRoute.isConnected was derived from RouteInfo.connectionState,
which stays DISCONNECTED for local SYSTEM routes even when active — so a
connected Bluetooth device showed "Available" and reported connected:false.
The picker subtitle now uses isSelected (route == selected route); the dead
isConnected field is removed and the misleading `connected` field dropped
from the diagnostics route event (it only ever logs the active route).
Refs Scribe M9 (#119), tasks #1204#1205.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
The diagnostics view already defaulted to the most recent 500 events (no
window); make that the obvious path. Device/Kind stay primary; the
start/end window + row cap move into a collapsed "Advanced filters"
disclosure (auto-opens when a window is active) with a "Reset to recent
500" action. Caption now states whether you're seeing the recent default
or a windowed slice.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
- DiagnosticsReporter.collectServerHealth: drop distinctUntilChanged() on
networkStatus.state (StateFlow is already distinct; the deprecation
warning is a hard error under allWarningsAsErrors).
- web users.test.ts: add debug_mode_enabled to the alice/bob AdminUser
fixtures now that the field is required on the type.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
drainSafe/drain each had 3 returns (detekt ReturnCount ≤ 2). Collapse the
guard clauses and convert drain's loop to a `more` flag — same behavior,
zero/two returns.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
Android: a gated DiagnosticsReporter taps connectivity, server-health,
UPnP drops/player-state/route, power (Doze/battery-opt/screen), and
app fg/bg, plus a heartbeat snapshotting Sonos-vs-local position — the
locked-phone desync signal. Events buffer in a Room ring buffer
(deliberately NOT the MutationQueue: high-volume best-effort telemetry
that must survive the dead zone being debugged) and DiagnosticsUploader
drains them on a tick / health-recovery / sign-in.
Gating: the account flag (users.debug_mode_enabled) reaches the device
via a new /api/me refresh in AuthController; a per-device local OFF
switch lives in Settings. Reporter runs only when enabled && !optOut;
disabling drops the unsent buffer.
Web admin: /admin/diagnostics — pick account+device+kind+time-window,
see a chronological timeline, flip an account's debug mode remotely, and
Copy-JSON / Download-NDJSON the slice for analysis.
Room schema 6→7 (new diagnostic_events table + auth_session.diagnosticsOptOut;
pre-v1 destructive fallback).
Refs Scribe M9 (#119), tasks #1174#1175#1176#1177.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
New diagnostic_events table + per-account users.debug_mode_enabled flag.
When an account's flag is on, its client(s) POST a batch timeseries of
connectivity / UPnP-sync / power / lifecycle events to /api/diagnostics
(no-op 204 when off, kind whitelist mirrors the CHECK constraint).
Admin surface: GET /api/admin/diagnostics (optional account/device/kind/
time-window filters, RFC3339-or-epoch-ms, export-sized paging) + a
/diagnostics/devices overview + PUT /api/admin/users/{id}/debug-mode to
flip an account remotely while a bug is live. debug_mode_enabled is now
exposed on /api/me (client gate) and the admin user views.
Retention: a 30-day gc-worker sweep (GcPruneDiagnostics), keyed on the
server clock so a skewed device clock can't keep rows alive.
Refs Scribe M9 (#119), tasks #1172#1173.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K55iTxn95BtshocgdE1shW
#980, parity with web e932ab43. When playlist.system_rebuilt arrives (SSE)
while a system-playlist detail screen is open, the ViewModel marks it stale and
the screen shows an indefinite "This mix was refreshed · Refresh" snackbar.
Refresh re-resolves the rotated variant via PlaylistsRepository.systemShuffle
and reuses the existing regenerated navigate-replace flow to land on the fresh
playlist id — without triggering another server rebuild (unlike the manual
regenerate button). Dismiss clears the flag. Functional behaviors were already
correct; this closes the cosmetic stale-list gap.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
#980. When the daily rebuild fires while a system-playlist detail page is
open, its cached data goes stale and can't be refetched in place — the
playlist id rotated, so the old id 404s. serverEvents now exposes a monotonic
rebuild counter; the detail page shows a "this mix was refreshed" banner with
a Refresh that re-resolves the variant (systemShuffle) to the new playlist id
and navigates there. No forced redirect, no auto-reload — the user refreshes
on their terms. Functional behaviors were already correct (tapping a song
plays it; tiles load the current mix); this closes the cosmetic list-staleness.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Playing a system playlist from /playlists/<id> previously sent no source, so
it never advanced that playlist's rotation — inconsistent with the home tile
(and the Android detail screen, which already tags the variant). Pass
source: variant alongside the existing self-heal closure so a play is
attributed regardless of the surface it started from. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Parity with the web SSE consumer (5a80a1e4). HomeViewModel now subscribes to
EventsStream and re-pulls Home (refreshIndex + system-playlist status) when
the server emits playlist.system_rebuilt — the daily 03:00 rebuild or a
manual refresh — so the system-playlist tiles and You-might-like rows reflect
the new snapshot without a manual reload. Browse-only: the active playback
queue is left to self-heal on the failure path. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The web client only ever SENT events; it had no inbound SSE listener, so a
tab left open across the daily system-playlist rebuild kept showing
yesterday's home + playlist snapshots until a manual reload (the stale-
browse-view bug behind #968). Add useServerEvents(): opens /api/events/stream
while authenticated and, on playlist.system_rebuilt, invalidates the home,
playlists, and system-playlist-status query caches. Deliberately does not
disturb the active playback queue — that self-heals on the failure path.
Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The daily 03:00 scheduler rebuild (and the manual refresh endpoint) replace
a user's system playlists + You-might-like rows but published no event, so a
client left open across the rebuild served yesterday's snapshot until a
manual reload — the stale-tab case behind #968. Add a user-scoped
playlist.system_rebuilt event (envelope {kind,user_id,data:{}}) from both the
scheduler (bus threaded into NewScheduler) and handleSystemPlaylistRefresh.
Clients consume it to invalidate home / system-playlist views and proactively
re-pull a stale active queue. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Web parity with 27766ae0. When a load error exhausts a fully-unplayable
queue, re-pull the source instead of stopping: a bare-variant source is a
refreshable system playlist (re-pull via PlaylistsRepository.systemShuffle),
"radio:<seed>" re-seeds via RadioController. Reads the source from the
current MediaItem extra; bounded to one re-pull per exhaustion (reset when
a track next loads with real audio) so a still-stale refresh can't loop.
Album / artist / user-playlist / offline sources have nothing to refresh
and still stop. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
When the whole queue proves unplayable (e.g. a tab left open across the
daily system-playlist rebuild — the exact stale-snapshot case), the player
now re-pulls the fresh snapshot and resumes instead of dead-ending on
"Try again". The seeder hands the store an opaque refetch closure so the
store stays decoupled from the playlist API and the per-artist
(songs_like_artist) identity problem: single-instance variants re-pull via
systemShuffle, per-artist mixes via getPlaylist(id), radio re-seeds from
its track. Bounded to one self-heal per exhaustion (reset on the next
successful play) so a still-broken refresh can't loop; "Try again" stays
the genuine last resort. Wired from PlaylistCard, the playlist detail page,
and playRadio. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
onPlayerError fired a `load_failed` event and the snackbar reporter coalesced
it into "Skipped N unplayable tracks" — but nothing actually skipped, so a
bad/stale track stranded playback while the toast claimed otherwise. Mirror
the zero_duration path: advance to the next item and re-prepare (a load error
leaves the player IDLE), or stop at the end. Forward-only bounds a fully-
unplayable queue. Web parity with 2a8de82a. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A track that fails to load (e.g. a stale system-playlist snapshot pointing
at a rebuilt/removed file) hard-set the player to the 'error' state and
stranded the user on a "Try again" button that just re-queued the same
failing track. Now a load error advances to the next track; the error
state only surfaces once the whole queue has proven unplayable — every
track failed, or we reached the end. A failure streak capped at queue
length stops a fully-broken queue from cycling, and resets on the next
successful play.
Next (Track B cont.): self-heal a stale system-playlist / radio queue by
re-pulling the fresh snapshot on total failure, plus the Android
equivalent. Issue #968.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Cover tiles (worst in the "You might like" home row, which surfaces
unplayed items whose art is often not yet backfilled) sat empty while
loading and stayed blank on a 404. The server returns a fast 404; the
gap was missing client-side loading/fallback states.
Web: new shared Cover.svelte owns the loading placeholder + onerror
fallback (static cover, or Disc3 for artists). AlbumCard, ArtistCard and
CompactTrackCard now reuse it instead of three hand-rolled <img> tags
that disagreed on fallback handling — notably ArtistCard had no onerror.
Android: ServerImage tracks Coil's load state so the per-caller fallback
doubles as a placeholder (loading) and an error state (404 / unreachable),
instead of only guarding the null-URL case. All five call sites pass an
explicit size modifier, so the new Box wrapper is layout-safe.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Make a fresh install usable out of the box and document the first-run flow
for the public-facing repo.
- Default scan_on_startup to true (Default() + config.example.yaml, which is
the live config baked into the image). Previously false, so a fresh stack
came up with an empty library and no hint to scan. Scans are incremental
(mtime skip), so the per-restart cost is just a directory walk. Re-point
the env-override test to exercise the override against the new default.
- README: add a "First run" walkthrough (register -> scan -> integrations ->
install Android app -> invite users), each grounded in a real route.
- Add docs/screenshots/ with six captures, referenced via width-constrained
<img> wrapped in a link (shrink inline + click to open full size).
API token and invite token were cropped/redacted out of the captures
before commit so no live credential lands in the public history.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bring the README and client/README in line with the current product and
release pipeline ahead of making the repo public-facing:
- Highlights: replace the retired "Flutter client in flight" bullet with
the native-Android-APK-bundled-in-the-image story.
- Quickstart: document the purpose of each compose volume mount, and fix
the minstrel-data mount (/data -> /app/data) so it matches the image's
MINSTREL_STORAGE_DATA_DIR and generated artefacts actually persist.
- Configuration: tie MINSTREL_STORAGE_DATA_DIR to the /app/data mount.
- Updating: replace the bogus :v1.0.x scheme with the real tag model
(:latest, immutable :vYYYY.MM.DD, :main) and note the bundled APK.
- client/README: rewrite the Production section to match release.yml
(android-release + needs: ordering, non-tag :latest bundle path).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Every main push moves :latest, but main builds don't build an APK — so the
in-app update channel silently vanished from :latest until the next tag.
Now the image-release job, on non-tag builds, pulls the most-recent
release's signed APK from the gitea API and reconstructs its exact
versionName (${TAG#v}.$(git rev-list --count TAG) — the same formula
android-release bakes in) for the version sidecar. No rebuild, just
rebundle; tag builds still bundle their own freshly-built APK. Checkout
gains fetch-depth:0 + fetch-tags so the commit count resolves. Degrades to
an empty client/ (404 update channel) — never a wrong version — if no
release / APK asset / tag count can be resolved.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
No single-click destructive action belongs in the kebab. Removing the item
orphaned its whole path (RemoveTrackPopover was its only caller, and the
admin/tracks API client was the popover's only caller), so per the repo's
no-dead-code convention the chain is fully removed: the menu item + its
admin/isAdmin plumbing in TrackMenu, RemoveTrackPopover(.svelte/.test),
src/lib/api/admin/tracks(.ts/.test), and the now-needless transitive mocks
in the CompactTrackCard / PlaylistTrackRow / playlist specs.
The kebab is now an 8-item, admin-agnostic menu. The DELETE /api/admin/tracks
server endpoint is untouched — a future safer admin surface can rebind it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The kebab gained "Start radio" and dropped the duplicate "Flag this track…"
(its action now lives solely under "Hide", which opens the same FlagPopover).
Net item count is unchanged (9 admin / 8 non-admin), but the named-item and
flag-entry assertions needed updating:
- mock playRadio in the store mock; assert Start radio dispatches playRadio.
- swap the flag-item presence check for start-radio.
- replace the "click Flag" test with "click Hide opens the popover".
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bring the web TrackMenu to parity with Android's canonical TrackActionsSheet
so the kebab reads the same on both clients:
- Reorder to Android's groups: queue → like/add-to-playlist/start-radio →
go-to-album/artist → hide.
- Drop the duplicate "Flag this track…" item — it opened the very same
FlagPopover as "Hide" (Android folds flag into a single Hide).
- Align icons (ListVideo / ListMusic / ListPlus / Disc3 / User).
- Admin-only "Remove from library" stays as a web superset (Android has no
surface for it), past its own divider.
Mount the kebab on the full-screen /now-playing route with hideQueueActions,
mirroring Android's NowPlayingScreen — Start radio / Add to playlist / Hide
were previously unreachable there (only like + volume + queue existed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Radio was fully wired (playRadio → /api/radio + 80% auto-refresh) but its
only entry point was TrackRow's inline 📻 button, so it was unreachable from
the kebab — i.e. missing on the Most Played compact cards and the mini-player.
Add a "Start radio" item to TrackMenu, shown even under hideQueueActions since
reseeding a station from the current track is meaningful there.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CompactTrackCard is a short one-line row but reused CardActionCluster,
which corner-splits Like+Add (top) and the kebab menu (bottom). That split
is right for the tall square Album/Artist cards but makes the two groups
collide on the compact row's hover state. Give the compact card a single
inline, vertically-centred right cluster (Like + Add + menu in one group)
and widen its right padding reserve to match.
In the desktop PlayerBar, the left info column was a fixed w-72 (title kept
truncating) while the seek column was flex-1 (the scrubber hogged the slack
on wide screens). Let the left column grow up to max-w-md while holding its
288px floor at md, and cap the seek/transport column at max-w-xl centred, so
freed width flows to the title instead of stretching the bar.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Three related improvements to UPnP/Sonos session handling, on top of the
WiFi-lock + drop-suppression fixes.
1. Adopt a running session instead of clear+reload (the headline).
Selecting a renderer always did removeAllTracksFromQueue + full reload --
a jarring restart if the speaker was already playing our queue (e.g. after
the phone got disconnected but the autonomous Sonos kept going). selectUpnp
now probes the renderer first; if it's mid-playback on the same track id at
the same queue index, we ATTACH in place: sync the local cursor to its
position, wire ActiveUpnpHolder, start polling -- no clear, no reload, and
skip/seek immediately drive its live queue. Falls through to clear+reload
when it isn't our queue. New PlayerController.moveCursorTo aligns the local
cursor without auto-playing.
2. Discovery expiry + selection revert (anti-stickiness). upsertRoute only
ever added, so a powered-off renderer lingered in the picker forever and
could pin a stale selection. Stamp lastSeen per route; after the picker's
active M-SEARCH scan, prune routes that didn't re-announce. A collector
reverts the selection to the phone when the selected route leaves discovery
while we're not actively casting -- so a later play never targets a ghost.
Pruning is tied to picker-open scans only (no background timer -> no row
flicker).
3. Reconcile immediately on network recovery. When NetworkStatus flips back to
Healthy while a route is active, nudge an immediate poll instead of waiting
up to POLL_INTERVAL_MS -- the held session re-confirms the renderer in one
round-trip.
Verified (read-only): the tap-play-onto-dead-route fallback still fires when
the phone's network is Healthy (the poll-loop drop path is unchanged for that
case); the drop-suppression gate only holds during phone-side outages, where a
local fallback couldn't play either.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Observed on device: casting to Sonos on battery + screen locked, a
transient ~67s reachability gap (NetworkStatus -> ServerDown while WiFi
itself stayed associated) starved the 1 Hz poll past DROP_THRESHOLD. The
poll loop then dropped the route and fell back to the local player, which
honored the play-intent -- so the phone suddenly started playing the song
out loud locally while the Sonos was still happily streaming it.
A poll failure during a phone-side network outage means "we can't see the
renderer right now," not "the renderer died": a UPnP renderer streams
autonomously and keeps playing, and the local player we'd fall back to
can't reach the server either. Dropping is strictly worse than waiting.
Gate the drop on NetworkStatusController: only drop when the phone's
network is Healthy (renderer genuinely unreachable on an otherwise-fine
link). While Unstable/ServerDown/Offline, hold the route, keep polling,
and clear the failure streak so recovery re-evaluates from scratch rather
than re-dropping on the first post-recovery hiccup. The poll reconciles to
the renderer's real (advanced) position once the network returns.
Complements the CastNetworkLock fix: the lock reduces how often these gaps
happen; this stops a gap that does happen from punishing the user.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
While casting, the wrapped ExoPlayer is paused, releasing its
WAKE_MODE_NETWORK locks -- so nothing kept the phone's radio awake. On a
locked, on-battery phone the WiFi power-saves within seconds and the CPU
dozes, stalling the 1 Hz liveness poll to the renderer and the
queue-extend calls to the server. The poll then trips DROP_THRESHOLD and
playback falls back to a phone that also has no network: silence, while
the Sonos was streaming fine the whole time.
Diagnosed from logcat: ~12s after screen-off the phone logged
"Unable to resolve host minstrel.fabledsword.com" (its own DNS, not the
server), the extend aborted (1/58 appended), then the drop tripped and
the local fallback came up active=null. USB charging masks it (no Doze
while charging), which is why it only bit on battery.
Add CastNetworkLock: a high-perf/low-latency WifiLock + partial WakeLock
acquired when a UPnP route goes active and released on drop/switch-back
(every teardown path funnels through holder.set(null) -> onActiveChanged).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per-source play outcomes so the operator can see whether each recommendation
surface is landing and tune the now-operator-tunable taste weights.
Server:
- query RecommendationSourceMetricsForUser: groups the user's play_events by
source (system-playlist surface), reporting plays / skips / avg completion
over a window; NULL-source (library/radio) plays excluded.
- GET /api/me/recommendation-metrics?days=30 (default 30, capped 365) →
{window_days, sources:[{source, plays, skips, skip_rate, avg_completion}]}.
- handler test: 401 unauth; per-source aggregation + NULL-source exclusion +
skip_rate / avg_completion math.
Web:
- lib/api/metrics.ts: query + friendly source labels.
- settings page gains a "Recommendation metrics" card (table of surface / plays
/ skip rate / avg completion), with loading/error/empty states.
- settings tests mock the new query (manual subscribe-store, hoisting-safe).
Note: You-might-like plays aren't source-tagged (it's a Home row, not a system
playlist), so this covers For-You / Discover / the mixes. Tagging YML plays
would be a client follow-up.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2a re-ranks the existing pool by TasteMatch; this ensures taste-relevant tracks
ARE in the pool. Adds a 6th arm to LoadRadioCandidatesV2: in-library tracks by
the user's top positively-weighted taste-profile artists ($10 K, weight > 0,
deterministic weight-DESC,id order so it doesn't reintroduce same-day
nondeterminism). Pool-inclusion only (sim_score 0) — TasteMatch already scores
the fit. Empty for cold-start users (no profile).
- CandidateSourceLimits.TasteOverlap; default 20 (radio), 80 for For-You via
systemForYouSourceLimits.
- You-might-like deliberately sets TasteOverlap=0: it surfaces NOT-actively-
engaged artists, so flooding its pool with top-taste (mostly already-played)
artists would just feed the read-time dedup.
- Test: positive-weight artist's track enters via the arm; negative-weight one
is excluded (weight > 0). Existing pool tests unaffected (no profile seeded).
Deferred within 2b: profile-seeded For-You — marginal given the arm + TasteMatch
already inject taste broadly (top-played seed ≈ top-taste artist).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The fallback pulled artists only from explicit artist-likes (general_likes_artists),
but most users like albums and tracks far more than artists — so the artists row
still came up thin (a couple of tiles) even with a rich library, while the albums
row filled fine.
Broaden both fallbacks to "entities you've shown affinity for":
- artist fallback = explicit artist-likes ∪ artists of liked albums ∪ artists of
liked tracks.
- album fallback = explicit album-likes ∪ albums of liked tracks.
New dedicated queries (ListYouMightLike{Artist,Album}FallbackForUser) replace the
narrow Rediscover-fallback reuse; same projection so the Go layer still converts
directly. (Aliased + fully-qualified the UNION arms — sqlc merges UNION scopes,
so unqualified user_id was ambiguous across the three like tables.)
Test: 12 liked TRACKS by distinct artists, no artist-likes → the artist row now
fills from their artists (was empty before).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
golangci-lint v2 (CI-only; local is v1) flagged the field-by-field struct
literals — the fallback and you-might-like row types are identical, so convert
directly instead.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The taste roll-up surfaces top-similar albums/artists, which for a heavy
listener are mostly ones they already play — so the read-time dedup (vs Most
Played + Rediscover + Last Played) can strip the section down to a single tile
(reported on the artists row). The code was sound; the section was just starved.
Adds a read-time fallback: when a You-might-like row comes up short after dedup,
top it up from the user's LIKED artists/albums — a far larger pool than the
12-entity similarity roll-up, so the same exclusions still leave plenty. Reuses
the existing Rediscover-fallback queries (no new SQL), applies the same
exclusions (already-shown + Rediscover + Most/Last Played) so it never
duplicates a tile or suggests an actively-played entity, and is best-effort
(a query error leaves the section as-is). Takes effect immediately — no rebuild.
A cold-start user with no likes gets nothing from the fallback, so the
new-user-empty behaviour is preserved (test still passes).
Test: 20 liked artists, none played → Rediscover fills 10, You-might-like
fallback fills the other 10, disjoint.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The web UI rendered a fixed set of Home sections and had no code for the
server's you_might_like_albums / you_might_like_artists (shipped in
v2026.06.11), so the row was absent in the web client. Adds it, mirroring
the Rediscover block, positioned directly under the system-playlists row.
- types.ts HomePayload: two new slices (server always emits them; web ships
in lockstep with the server).
- +page.svelte: a "You might like" section (albums + artists scrollers) as the
first section under the playlists row, with a "still learning your taste"
empty state for the cold-start/gated case. Reuses existing AlbumCard /
ArtistCard / HorizontalScrollRow.
- home.test.ts / page.test.ts: mock payloads gain the two fields.
Completes the You-might-like row across all three clients (server already
emits it; Android in v2026.06.11; web here).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Operator expected the row immediately beneath the system-generated playlists
section, not below Rediscover. Reorders the section call (presentation only —
no logic/state change) and updates the doc comment.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The two new you-might-like observe accessors pushed HomeRepository from 11 to
13 functions, tripping detekt's per-class default. It's accessor density (one
observe method per Home row) on a thin pass-through repository, not complexity —
suppressed with a one-line rationale per the project convention. ktlint already
passed; this was the only detekt finding.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Surface the server's you_might_like_albums / you_might_like_artists sections
(daily-built, cold-start gated, taste-aware) on the Home screen, mirroring the
Rediscover block.
- HomeIndexWire: two new slices, defaulted to emptyList() so decode is safe
against older servers that don't emit them (Class-B discipline).
- HomeRepository: two section constants + observeYouMightLikeAlbums/Artists
(reusing the existing album/artist hydration helpers) + refreshIndex now
replaces both sections and pre-warms their artists. No Room schema change —
cached_home_index stores the section string verbatim.
- HomeScreen: HomeSections gains the two fields (+ isAllEmpty); the ViewModel
combine is split (core 5 → +2 you-might-like → +playlists) to stay within the
coroutines 5-arity limit; a YouMightLikeBlock + youMightLikeSection render an
albums-then-artists block below Rediscover, with a "still learning your taste"
empty state for the cold-start/gated case.
Server side already shipped in v2026.06.11; this makes it visible on device.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The profile built in phase 1 now changes what gets surfaced. Adds a TasteMatch
term to the weighted-shuffle score so candidates are re-ranked by their fit to
the user's learned taste (positive draws toward it; negative reflects passive
avoidance; 0 at cold start).
- recommendation/score.go: ScoringInputs.TasteMatchScore ([-1,+1]) +
ScoringWeights.TasteWeight + the term in Score.
- recommendation/taste.go: LoadTasteProfile reads the taste_profile_* tables;
TasteProfile.Match blends the candidate's artist weight (0.7) and avg genre-tag
weight (0.3), each tanh-squashed by a fixed scale so one outlier artist can't
compress the rest. Unknown artist/tags and empty profiles → 0 (neutral).
- candidates.go: both candidate loaders set TasteMatchScore per candidate, so
every Score caller (system playlists incl. You-might-like, radio) becomes
taste-aware automatically.
- weights: systemMixWeights.TasteWeight = 1.5 (daily mixes are the primary
taste surface); config.RecommendationConfig gains taste_weight (default 1.0,
lighter — radio is seed-directed) wired into the radio handler.
- tests: pure (Match curve incl. saturation/clamp/empty-neutral, Score term
add+subtract) + DB round-trip (seed taste rows → Match positive). All green
vs real Postgres; existing playlist/radio tests unaffected (empty profile →
zero taste effect).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
golangci-lint v2 (CI-only; local is v1) flagged two unused-parameter issues:
- BuildTasteProfile's `now` was genuinely dead — decay/windowing are computed
DB-side via now(), so no Go-side timestamp is threaded. Removed it (a
phase-3 context model that needs a pinned reference time would re-add it);
updated the scheduler call site.
- the degenerate-params engagement test ignored t; reworked it to assert the
result stays in [-1,1], which also strengthens the test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Build a persistent, decaying model of each user's taste, recomputed daily,
that later phases consume across every recommendation surface. Phase 1 only
BUILDS the object — no behaviour change to what's surfaced yet.
Core mechanic — graded engagement (replaces binary was_skipped for learning;
was_skipped stays for History): a play's completion ratio maps to a signal in
[-1,+1] via two linear ramps (instant-skip → -1, ~0.30 neutral, ≥0.90 → +1).
Time-decayed (half-life ~75d) so recent behaviour dominates and the profile
tracks drift.
Per operator constraints:
- No explicit dislike button — negatives come only from passive behaviour
(early skips). Nothing recorded to regret or opt out of.
- Negatives are track-scoped; artist/tag weight is the decayed SUM of their
tracks' engagement, so one skip nets out against many good plays (a
DB test asserts a liked artist stays positive despite an early-skipped
track). A floor clamp bounds how negative any single entity can get.
- migration 0035: taste_profile_artists / taste_profile_tags (signed weight,
indexed by (user, weight DESC)).
- internal/taste: engagement.go (pure curve + decay) + profile.go
(accumulate plays + like bonuses, floor damping, size caps, atomic-replace).
- scheduler: rebuildUserDaily recomputes the profile before the playlist
build (so phase 2 can read it), best-effort — a taste failure never blocks
playlist building. Wired into the daily job + startup catch-up only (not
manual/lazy rebuilds).
- tests: pure (engagement curve, decay, ranking, floor, genre split) +
DB-backed (positive/negative weights, aggregation-protects-artist, like
bonus, atomic replace). All green vs real Postgres.
Config knobs live in taste.DefaultConfig() for now; wiring them into the
server RecommendationConfig is a later follow-up.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
scoreAndSortCandidates drew per-candidate jitter by slice position, but
the candidate query (LoadRadioCandidatesV2) has ORDER BY random() arms and
no stable outer ordering, so DB row order varies call-to-call. When the
recency spread between candidates is smaller than the ±jitter (small or
recency-clustered libraries), two same-day rebuilds assigned jitter to
different tracks and reordered near-ties — so the build was not actually
deterministic-within-a-day as documented.
Pre-existing latent flake in TestBuildSystemPlaylists_DailyNonceDeterminism
(passed in isolation / by luck in CI; deterministically reproduced when the
system-build tests run in sequence). Confirmed independent of the
You-might-like change by neutralizing buildYouMightLike — the flake
persisted.
Fix: sort the candidate slice by track id before assigning jitter, so the
jitter for a track is a function of (track, day) alone, independent of DB
return order. Verified: full playlists package green 4/4 and the build-test
sequence green 5/5 (was 0/4 before).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Surface in-library albums/artists the listener doesn't actively spin but
is predicted to enjoy, derived from the same similarity + like-weighted
candidate engine that powers For-You — rolled up from track scores to
album/artist granularity. Built in the daily 3am BuildSystemPlaylists
pass, atomic-replaced alongside the system playlists, and read back by
/api/home (+ /api/home/index).
Cold-start gate: skips generation entirely below 20 distinct unskipped
tracks AND 5 distinct artists, so a thin profile ships empty rows rather
than near-random tiles.
- migration 0034: you_might_like_albums / you_might_like_artists (id+rank,
CASCADE, per-user rank index).
- playlists/you_might_like.go: cold-start gate + similarity roll-up
(sum-of-top-3 aggregation, per-artist album cap, daily-rotating via the
same userIDHash jitter as For-You) + atomic-replace persist in the tx.
- recommendation/home.go: two new HomePayload sections with read-time
cross-section dedup vs Most Played / Rediscover / Last Played, trimmed
to 10 each.
- api: you_might_like_albums / you_might_like_artists on /api/home and
/api/home/index, reusing albumRefFrom / artistRefFromCovered.
- tests: pure roll-up/aggregation/cap unit tests + DB-backed gate,
sufficiency, and atomic-replace tests (all green vs real Postgres).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
LoopWithTooManyJumpStatements — replace the two continues with a
filtered sequence + a single null guard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
bvandeusen