Bumps base version 2.8.1 → 2.8.2 and the docker-publish default tag. A stability + performance
release: Spotify reliability (Docker boot-hang #949, the token-cache re-auth fix, on-demand
Sync-to-Spotify), the "slow after update" password-manager fix + Max Performance mode (#948), and
large-library imports that no longer time out the import page (#947).
Updates the five release touch-points: web_server version, docker-publish default, pr_description.md,
helper.js WHATS_NEW + VERSION_MODAL_SECTIONS (current release + Earlier-in-2.8.1 summary), and the
new RELEASE_2.8.2_discord.md (truncated for Discord, 1351 chars).
Re-review caught a real bug in the phase-2 polling: it refetched only the files query, but the
album import tab uses its OWN groups query (album-import-tab.tsx). So after a large-folder scan
completed, the album tab would stay stuck on its initial {scanning} response and never populate
(the singles tab was fine — it reads the polled files query). Switched the scan poll to
invalidateImportStagingQueries (files + groups + suggestions) so every mounted staging query
refetches when the scan finishes. (Suggestions is already async/cached, so it just no-ops.)
typecheck clean, scanning test still passes.
Frontend half of the async staging scan. The endpoints now return {scanning, progress} while a
large staging folder is still being scanned in the background; the page surfaces that and fills in
automatically when it completes — no manual refresh, no timeout.
- types: ImportStagingFiles/GroupsPayload gain optional scanning + progress (additive).
- useImportStaging exposes `scanning`/`scanProgress` and, while scanning, polls via a plain
setInterval(refetch, 1500). Deliberately NOT react-query's refetchInterval — and a plain interval
that only runs while scanning leaves the normal + error query states completely untouched.
- the header shows "Scanning N of M files…" instead of a count while the scan runs.
vitest: new test asserts the scan-progress header renders from a {scanning} response; typecheck
clean. Note: -route.test.tsx's pre-existing "staging files fail to load" test fails only in the
full-file run (passes in isolation) — verified it also fails on clean dev with all my changes
stashed, so it's a pre-existing test-isolation flake, not from this change.
Double-checking the on-demand auth flow: the needs_auth handler called window.open() AFTER an
await, which breaks the user-gesture chain so browsers popup-block it — the user would see "approve
in the new tab" with no tab. Replaced with a clickable authorize link (a direct click is never
blocked).
Adds two endpoint tests via the Flask test client: Spotify export returns needs_auth + the
/auth/spotify/export url (and short-circuits before the DB) when the token lacks write scope, and
does NOT short-circuit when write scope is present. 10 service-export tests green, 64 script-integrity
green, ruff clean.
Brings back Spotify playlist export WITHOUT the regression that forced every user to re-auth.
The safety property: the global login scope (SPOTIFY_OAUTH_SCOPE) is NEVER changed, so no
existing token is invalidated. The write permission is requested only when a user actually
exports to Spotify.
- SPOTIFY_EXPORT_SCOPE = the global read scope + playlist-modify, used ONLY by the new
/auth/spotify/export route. Spotify returns a superset token; the normal /callback exchanges
and stores it unchanged (read ⊆ read+write keeps the standard auth check valid) — no callback
changes needed.
- SpotifyClient.has_write_scope() checks the cached token for playlist-modify.
- start_playlist_export_service returns {needs_auth, auth_url} for Spotify when the token lacks
write, instead of starting a doomed job. The modal opens the consent in a new tab and tells the
user to retry once approved; the "Sync to Spotify" button is back, gated on connection as before.
- Release notes (pr_description / What's New / version modal / discord) restored to Spotify &
Deezer with the one-time-permission note; discord back under 2000 chars (1983).
Tests: export scope is a strict superset of the (still read-only) global scope; has_write_scope
true/false for write/readonly/missing tokens and no-client. 275 spotify/oauth tests green, ruff
clean, 64 script-integrity green.
Follow-up to the auth hotfix (633aa82b). The Spotify playlist-write scope was reverted out of the
global OAuth scope (it was force-invalidating every user's token on upgrade), so "Sync to Spotify"
can't get write access yet — clicking it would dead-end on a misleading "reconnect Spotify". So:
- removed the "Sync to Spotify" button from the export modal (Deezer stays); the backend write
client + endpoint are left in place, dormant, for when on-demand write-auth lands
- modal copy is now Deezer-only ("Match missing tracks (Deezer)", "stored Deezer ID")
- release notes (pr_description, helper.js WHATS_NEW + version modal, RELEASE_2.8.1_discord.md)
reworded from "Spotify & Deezer" to "Deezer", with a "Spotify export coming in a follow-up" note
64 script-integrity tests green; discord file back under the 2000-char limit (1952); no stale
Sync-to-Spotify mentions remain. Deezer export (live-verified) is unaffected.
The settings info icons are role="button" spans with a text "i" glyph but no
cursor/user-select, so hovering the glyph gave the I-beam text caret on Windows
(Linux happened to resolve a pointer). Add cursor:pointer + user-select:none so
it reads as a button on every platform.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Max Performance only neutralized animation/blur/shadow globally but didn't
replicate the reduce-effects-specific display:none rules, so with reduce-
effects OFF the sidebar aura orbs (.sidebar::before/::after) survived as two
hard static circles, the dash-card cursor-glow layers stayed, and nav-button
hover kept the expensive treatment. Depended on whether reduce-effects was on
before enabling Max Performance. Extend those three rule blocks to also match
body.max-performance — flash-free since the body class is server-rendered.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Expose window.__pmSuppress.disable()/enable() on the suppression IIFE so a
before/after benchmark can reproduce the pre-fix "before" state (managers
re-attach their autofill overlay) and restore it, without a rebuild. The
app itself never calls these; suppression stays on by default.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two CPU regressions surfaced in software-rendered / no-GPU containers
(Docker), where transform/opacity and canvas radial-gradient fills
rasterize on the CPU instead of a compositor:
1. Worker-orbs canvas + decorative motion saturate a core and freeze the
UI. A new opt-in "Max Performance" mode is the nuclear low-power switch:
body.max-performance CSS kills blur/shadow/filter AND all
animation/transition (spinners go static), and JS halts every canvas
loop (orbs, particles, cursor-glow, API sparks) via window._maxPerfActive.
Reduce Visual Effects is now decoupled from the orbs — they follow their
own toggle; only Max Performance force-kills them. While Max Performance
is on, the Orbs/Particles/Reduce-Effects checkboxes lock + grey out, and
save reads the runtime flags so prefs aren't clobbered.
2. Password managers (Bitwarden et al.) rebuild their autofill overlay on
every DOM mutation; a captured trace showed Bitwarden using ~6x the CPU
of the whole app (~400 setupOverlayOnField/sec). suppressPasswordManager-
Autofill() tags non-credential inputs with data-bwignore / data-1p-ignore
/ data-lpignore / data-form-type=other so the managers skip them; real
login/PIN fields are left alone.
Wired through: web_server.py (_initial_appearance_context), index.html
(inline flag + body class + checkbox), init.js (applyMaxPerformance +
bootstrap + listener + autofill suppression), settings.js (load/save),
worker-orbs.js / particles.js / api-monitor.js (gates), style.css.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Bumps base version 2.8.0 → 2.8.1 and the docker-publish default tag. Headline is the
Spotify/Deezer playlist export (#945); also the Library Reorganize rename-only mode (#875),
broader lossless handling (#941/#939), download + search fixes, the refined reduce-visual-effects
pass, and merged contributor PRs (#942/#943/#944).
Updates the five release touch-points: web_server version, docker-publish default, pr_description.md,
helper.js WHATS_NEW + VERSION_MODAL_SECTIONS (current release + Earlier-in-2.8.0 summary), and the
new RELEASE_2.8.1_discord.md.
The export modal now checks connection on open via /api/discover/your-albums/sources (cheap
token/ARL check, no live verify) and greys out + relabels any service that is not connected
("Not connected — set up X in Settings → Connections first"). Clicking a gated button nudges to
Settings instead of starting an export that would just fail with "not connected". The fetch runs
after the modal renders, so a slow/failed check never blocks the modal (buttons stay usable).
Pairs with the existing scope-403 handling: a Spotify token without playlist-modify still shows as
connected (it IS), and the writer returns the clear "Reconnect Spotify to grant playlist write
access" message — so "not connected at all" and "connected but needs reconnect for write" are both
covered. Static file, no rebuild.
Adds the third resolver stage for tracks the discovery cache + library can't resolve — a live
search of the target service, gated behind a "Match missing tracks" toggle so the API cost is opt-in.
The whole point is coverage WITHOUT the wrong-track risk, so it's a CONFIDENT match, not "search
and grab":
- search_service_track_id(artist, title, search_fn): searches the service, reranks via the existing
relevance scorer (filter_and_rerank), and returns the top hit's id ONLY if it clears
BACKFILL_MIN_SCORE (1.2 on the score_track scale). A wrong-artist hit (no 1.5x exact-artist boost,
caps ~1.0) or a karaoke/cover (x0.05) can't clear the floor → None, and the track is left out
rather than added wrong. search_fn injected → unit-testable without a live service.
- resolve_service_track_ids gains an optional search_id_fn: cache → library → search. Tallies
from_search separately.
- _run_service_export builds the search fn from the service's metadata search client only when
job['backfill'] is set; the endpoint reads `backfill` from the body; the modal adds the toggle and
the status line shows "(N matched live)".
Store-back of confident matches deferred: a mirrored-only track may have no library row to write to,
so persisting needs the track→library mapping — a follow-up, not correctness.
9 new tests incl. the safety ones: wrong-artist rejected, karaoke/cover rejected, real-over-cover
picked, fail-safe on search error, and the cache→library→search waterfall + toggle wiring (on/off).
28 export/orchestration tests green, 64 script-integrity green, ruff clean.
Adds "Sync to Spotify" and "Sync to Deezer" buttons to the mirrored-playlist export modal
(#pl-export-modal), alongside the existing ListenBrainz/JSPF options. They POST to the new
/export/service/<service> endpoint; the shared poller now reports service progress
("Pushing to Spotify…"), the matched/unmatched count, and links the created playlist on
done. ListenBrainz/JSPF paths unchanged. Static file, no rebuild.
This completes the #945 vertical: resolver → Spotify+Deezer write clients → export job +
endpoint → modal. Reverse-sync to a service is a clean export (uses the stored per-track
service IDs from enrichment), distinct from true bidirectional sync.
The Wing It pool "Fix Match" search returned "no results" for everything (even obvious
tracks). Root cause: /api/spotify/search_tracks built a Spotify field-filtered query
(track:X artist:Y) and handed it to spotify_client.search_tracks, which falls back to the
user's configured source when official Spotify isn't serving the request. The fallback
(Deezer here) got the raw Spotify `track:…artist:…` syntax it can't parse and aborted the
connection (RemoteDisconnected) — so the user's perfectly working Deezer failed ONLY on
this path, on this query format. The iTunes and Deezer search endpoints already dropped
field syntax for exactly this reason; the Spotify one was the lone holdout.
fix:
- new pure helper relevance.build_combined_search_query(track, artist, legacy) — plain,
source-agnostic query; documents WHY field syntax is wrong here. the endpoint already
reranks by expected title/artist, so precision is recovered without the brittle syntax.
- the Spotify endpoint uses it (now consistent with iTunes/Deezer).
- frontend (searchPoolFix): surface the real error (auth / 500 / upstream abort) instead
of masking everything as "No results found" — which is what made this undiagnosable.
5 helper tests incl. the regression (output must contain no 'track:'/'artist:' syntax).
654 metadata/search tests green, 64 script-integrity green, ruff clean.
Adds an "Action" selector to the reorganize modal — "Full reorganize (default)" vs
"Rename only (skip post-processing)" — with a hint explaining rename-only skips
re-tagging/quality/AcoustID, only touches files whose name changes, and that renaming
can reset media-server play counts / date-added. executeReorganize sends rename_only in
the apply POST. Default is full → existing behaviour unchanged. Static file, no rebuild.
Pairs with the previous commit (orbs now run under reduce-effects). When the user has asked
for performance (reduce-effects on) we don't need the orbs at 60fps — the slow drift and sparks
are indistinguishable at 30, and dropping every other render roughly halves the per-frame canvas
cost, keeping the "orbs under reduce-effects" experiment cheap.
The canvas still ticks at 60fps and frameCount still increments every tick, so `time` stays
real-time and the drift speed is unchanged — we just draw it half as often. Precedence: the
existing fully-asleep ~20fps throttle still wins; the 30fps cap only applies awake + reduce-effects.
Chrome users with full effects keep 60fps — no reason to dim them.
Reduce-effects used to force-kill the worker orbs (isEnabled() had && !_reduceEffectsActive),
which also made the orb toggle a dead setting whenever reduce-effects was on.
The assumption was "the orbs ARE the expensive thing." On inspection that looks wrong: the
dashboard orb glow is drawn with canvas radial gradients, not a CSS blur(28px). The genuinely
expensive blur is the SIDEBAR aura orbs + frosted glass (CSS filters), which reduce-effects
still kills via filter:none regardless. So the orb canvas's per-frame cost should be moderate,
not the blur-rasterize lag.
So decouple them: the worker-orbs toggle controls the orbs on its own; reduce-effects keeps
killing the expensive CSS rendering but no longer gates the orbs. This also fixes the dead-toggle
conflict (the orb toggle now works under reduce-effects instead of being silently overridden).
Empirical: try it and watch the dashboard CPU. If the orb canvas under reduce-effects pushes it
back up, revert is one token — re-add `&& !window._reduceEffectsActive` to isEnabled().
The blurred 60fps worker-orb canvas is the main remaining Firefox lag source after the
#935 sweep (multiple Discord lag reports). So for a FIRST-TIME user with no saved
preference, default the orbs OFF on Firefox (smooth first impression where it's needed)
and ON everywhere else (full polish where the browser handles it). An explicit saved
choice ALWAYS wins — this only picks the default when the user hasn't chosen.
Done kettui-style with a SINGLE source of truth, not the dual browser-detection I first
floated (server UA + client _isFirefox would be the same fact in two places that can
drift — exactly the server/client class #943's green-flash fix just cleaned up):
- core/ui_appearance.py (new, pure + importable): is_firefox_user_agent +
resolve_worker_orbs_default(explicit, is_firefox) — explicit wins, unset → !firefox.
- web_server: the SERVER decides (UA via _request_is_firefox, request-context-safe) and
injects initial_worker_orbs_enabled; config default flipped None so "unset" is
distinguishable from an explicit False. The client just consumes the injected value
(init.js unchanged) — no client-side re-derivation of "is Firefox".
- settings.js: the orb checkbox default now reflects the server value when unset, so
saving Settings can't silently flip a first-time Firefox user's orbs back on.
No regression: Chrome users unchanged; users with an explicit setting unchanged (it
wins regardless of browser); /api/settings returns raw config so it can't clobber the
default for an unset value. Verified end-to-end through a real Flask request context
(Firefox→off, Chrome→on, explicit wins both ways, no crash outside a request). 8 pure
seam tests pin the contract; ruff clean.
"Reduce visual effects" was a sledgehammer: body.reduce-effects * forced
animation:none + transition-duration:0s on every element. That froze every CSS loading
spinner mid-rotation — including the dash-header worker-service spinners (musicbrainz /
spotify / deezer / … .active .<svc>-spinner) — which read as BROKEN rather than "off".
It also killed cheap hover feedback like the Quick Actions buttons.
The actual lag (esp. Firefox, see the #935 sweep) is backdrop-filter / box-shadow /
filter re-rasterizing every frame — NOT the animations themselves. Transform- and
opacity-only motion (the spinners) composites for ~free.
So: keep forcing the expensive properties to none (unchanged — that's the real fix),
but drop the blanket animation/transition kills. !important author declarations outrank
animation + transition declarations in the cascade, so any keyframe/transition that
tries to set blur/shadow/filter is still neutralized even while it runs — the spinner
spins, just without the glow. Net: functional spinners stay alive, Quick Actions hover
(transform + border-colour) returns, box-shadow transitions are no-ops (shadow forced
none), and the GPU-heavy rendering that caused the lag stays gone. The worker-orb CANVAS
is unaffected (JS-gated separately) and stays off under reduce-effects, as intended.
Static guard test pins the contract: the global rule must keep the expensive-property
kills and must NOT reintroduce blanket animation:none / transition:0s.
Discord (Shdjfgatdif): the import-singles Identify search prefilled "artist title" (space), so
"Sub Focus Last Jungle" returned junk while "Sub Focus - Last Jungle" found the track. The
placeholder already hints "Search artist - title..."; the prefill just did not match it. Join with
" - " instead of " ". filter(Boolean) keeps a lone title (no artist) dash-free.
diegocade1: DSD files (.dsf, ~500MB DSD64) were labeled "Low Quality" and nagged to upgrade.
two independent causes, both fixed (additive — no existing format/behaviour changed):
1) DSF was an unrecognized format -> bottom 'unknown' tier -> "Low Quality":
- source_map: map .dsf/.dff -> 'dsf' (also lights it up in AUDIO_EXTENSIONS, so Soulseek can
match a DSF if one exists)
- model.tier_score: 'dsf' base 102 (just above FLAC) — lands in the lossless range
- probe_audio_quality: add a DSD branch returning format='dsf' (mutagen.dsf for .dsf detail;
.dff classifies lossless without measured detail) instead of None
- settings UI: DSD in RT_LOSSLESS_FORMATS + a "DSD (DSF / DFF)" option in the profile dropdown
2) the actual cause of the screenshot's findings — the truncation guard falsely called DSF
"broken (only ~12% decodes)": ffmpeg decodes DSD to PCM at a different rate than the DSD
container's 2.8 MHz, so astats samples ÷ container-rate massively under-counts. now
detect_broken_audio skips the truncation check for DSD (silence detection still applies).
8 seam tests: dsf/dff -> 'dsf'; dsf tier in lossless range (with + without measured bitrate);
is_dsd_path; and a contrast pair proving the same 12%-decode numbers flag a .flac but skip a
.dsf. 230 quality/import/silence tests green, ruff + JS integrity clean.
since 9a0e3b40 persisted completed downloads in the Downloads view, the Clear Completed button
was hidden for those rows and clear-completed only pruned live session tasks. after a restart
the page filled with persisted completed downloads with no way to clear them.
now Clear Completed clears BOTH:
- live session completed/failed tasks (clear_completed_local, unchanged), AND
- the persisted download-history tail: new clear_completed_download_history() deletes every
library_history event_type='download' row, so the list actually empties and stays empty.
this includes unverified rows (the verification review queue) by design: on a library where
verification never confirmed the imports, ALL completed downloads are 'unverified', so preserving
them made the button a no-op. it only removes HISTORY rows — the actual files and their tracks
entries are untouched, so nothing in the library is lost, only the 'needs verification' flags.
the action confirms first (showConfirmDialog, destructive) and the button now shows whenever any
completed/failed row is present.
3 seam tests (clears all incl unverified; leaves non-download history; empty=0); reconcile +
orphan + JS integrity suites green.
the dashboard orb canvas kept falling back to ~1fps on firefox after the page settled. root
cause: firefox throttles requestAnimationFrame to ~1fps for a canvas it heuristically deems
occluded. the WAA keepalive only delayed the heuristic; it re-fired over time.
real fix: on firefox, drive tick() with a setInterval(~60fps) instead of self-scheduling rAF —
setInterval is not subject to the canvas-occlusion rAF throttle, so the orbs stay at full
framerate indefinitely. chrome is untouched (keeps vsync-aligned rAF). same render workload
(idle still drops to 20fps via the existing sleep skip); background tabs still parked by the
visibilitychange handler → stopLoop clears the interval. kept the keepalive (harmless).
root cause: the library stores album/artist art as media-server RELATIVE paths (Plex
/library/metadata/.., Jellyfin /Items/.., Navidrome /rest/..), which don't render in a browser
<img>. normal wishlist items carry Spotify CDN urls so they show fine, but LIBRARY-sourced
items — dead-file re-downloads and preview-clip re-fetches — carry the raw relative path, so
their album art came up blank. and the nebula only had artist photos for WATCHLISTED artists,
so non-watchlist orbs showed initials.
fix on READ in the wishlist tracks endpoint (so it also repairs items already in the wishlist,
no re-run needed), using the library data we already have:
- normalize each track's album.images url that needs it — relative/internal only, via the
canonical normalize_image_url; CDN urls are left untouched so already-rendering items can't
regress.
- build an artist-name -> normalized library-photo map and return it; the nebula seeds its
artist-image map from it (every wishlist artist), with curated watchlist photos overriding.
8 tests (predicate: relative/internal fixed, CDN untouched; album normalize in-place; artist
map build/skip-empty/idempotent/graceful). 237 wishlist+repair+JS tests green, ruff clean.
The reconcile heals rows whose file is still in the library; it deliberately
leaves ORPHANS — history rows whose file is gone (deleted / replaced /
re-downloaded elsewhere). Those can never be healed (no file left to confirm)
and linger in the Unverified list forever. This adds an explicit, user-initiated
cleanup for them.
- core/downloads/orphan_history.py: pure, tested rule. A row is an orphan when
its file resolves nowhere; flags `suspicious` when EVERY reviewed file is
unreachable (the mount-down signature) so the caller refuses rather than
mass-delete a healthy log during an outage.
- POST /api/verification/clean-orphans (admin-only): runs it against
_resolve_history_audio_path (raw path -> prefix-swap resolver -> tracks-table
title fallback), refuses on the suspicious signature, and deletes only history
ROWS — never a file (the files are already gone).
- UI: "🧹 Clean orphaned" button in the Unverified bulk-actions row, with a
confirm dialog spelling out that it removes log rows only and refuses if the
library looks offline.
NEVER automatic / never at boot — a filesystem check during a mount outage would
otherwise wipe good history. 5 pure-rule tests + safety-gate coverage.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LWJk7EuM7YktQeNyqQwTZY
each preview-clip finding now renders a dedicated detail card: File Length vs Real Length
(e.g. 28s vs 200s) and a Play button so the user can listen to the clip and confirm it's a
busted ~30s preview before approving the delete + re-download. reuses the existing
_renderPlayButton/playFindingTrack path that dead_file/orphan findings already use (the
finding carries file_path + title/artist/album, everything it needs).
HiFi (and occasionally other) downloads sometimes deliver a ~30s preview clip instead of the
full song; it lands in the library looking real. new repair job scans short tracks (duration
<= 30s, configurable), looks up the EXPECTED length from the track's metadata source
(spotify/itunes/mb get_track_details), and flags any whose real length is much longer than the
file (default: >= 30s longer) as a preview clip.
approving the finding (repair_worker._fix_short_preview_track) deletes the preview file (path
resolved via _resolve_file_path like the other delete tools), drops the DB row so the track
goes missing, and re-adds it to the wishlist with the full payload (mirrors _fix_dead_file)
so the real version downloads. scan ONLY creates findings — nothing destructive without user
approval, like every other tool.
conservative: genuine short tracks (source agrees they're short) and tracks whose length can't
be verified are skipped, never flagged. registered the job + finding-type label/fix-button in
the UI. 5 tests (scan flag/skip/scope + fix delete+remove+wishlist); 89 repair tests green.
the Memory Usage stat showed only global system memory (psutil.virtual_memory().percent).
add the process's own resident set size (RSS) — the real 'how much RAM SoulSync uses' number —
formatted MB under 1GB, GB above. headline stays the system %, subtitle now reads 'SoulSync ·
612 MB' instead of the generic 'Current usage'. graceful fallback if psutil errors / older
backend. useful context after the recent RAM-footprint discussions.
Firefox re-rasterizes blur()/backdrop-filter every composite where Chrome caches it, so the
always-visible shell glass (sidebar header + aura orbs, hero/header buttons) was ~half of
Firefox's idle GPU. gate behind @supports(-moz-appearance:none) so it's Firefox-only: hide the
two blur(28px) sidebar orbs + the dash-card blobs, and drop backdrop-filter on the sidebar
header and hero/header buttons (each keeps its tint, just unfrosted). measured ~20-25% -> ~10-13%
on Firefox, every page (sidebar is always visible). chrome is untouched — the block doesn't
exist there, full frost intact.
removing the always-on dash-card blob animation (for the Chrome GPU win) incidentally let
Firefox start throttling the worker-orb canvas's compositing to ~1fps after a header hover
re-layerizes the dashboard — Chrome never throttles it. re-add the 'keep the compositor warm'
effect cheaply: a 2px, ~invisible element running an infinite transform-only animation (zero
paint). gated behind CSS.supports('-moz-appearance') so it's Firefox-only; Chrome never gets
it. confirmed fix in Firefox/Zen.
the .dash-card cursor blobs are 16 large blur(48px)/blur(18px) layers. chrome caches them
once; firefox re-rasterizes blur on every composite, so they're a big chunk of idle dashboard
GPU on firefox. they're purely decorative and reduce-effects already hides them. gate behind
@supports(-moz-appearance:none) so it's firefox-only — chrome keeps the full cursor glow,
this block doesn't exist there.
