# Changelog

All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/),
and this project adheres to [Semantic Versioning](https://semver.org/).

## [Unreleased]

## [2.2.1] - 2026-02-09

### Added
- **Per-codec encoder recommended quality** - Quality sliders now show the encoder name and VMAF-calibrated default below each slider (e.g., "Recommended (VAAPI): 20")

### Fixed
- **Duplicate hardware scale filters** (#102) - Restructured encoder filter chain to eliminate duplicate HW scale filters that caused transcode failures
- **VMAF-calibrated quality defaults** - All encoder defaults recalibrated via VMAF binary search to target ~83 VMAF consistently across QSV, VAAPI, NVENC, and software
- **Per-codec encoder detection** - HEVC and AV1 now detect hardware encoders independently, fixing wrong defaults when codecs use different hardware (e.g., NVENC HEVC + software AV1)
- **Quality validation accepts auto mode** - Setting quality to 0 (use encoder default) no longer rejected by API validation

### Changed
- Collapsed identical main10 profile switch arms into single statement
- Updated stale comments referencing old 0.35 bitrate modifier values

## [2.2.0] - 2026-02-08

### Added
- **Browse page sorting** (#97) - Sort files by name or size in the browse view

### Fixed
- Preset dropdown badges (HW/SW) clipped on mobile portrait
- Queue stats dropdown clipped on left edge on mobile
- Browse sort labels and arrow convention aligned with queue menu

### Changed
- Removed Podman section from Quick Start (info remains in FAQ)

## [2.1.1] - 2026-02-07

### Changed
- **SmartShrink is now SDR-only** — Removed HDR VMAF support; HDR files are skipped when using SmartShrink presets. Use Compress presets or enable tonemapping for HDR content. This simplifies the VMAF pipeline and avoids edge cases with HDR tonemapping during analysis.
- **VMAF frame subsampling** — Scores every 5th frame instead of every frame during VMAF analysis, cutting scoring time ~5x with negligible accuracy impact (same technique used by ab-av1)
- **GitHub link in settings** — Added "View on GitHub" link next to version number in settings footer

### Fixed
- Refresh button stuck state when CSS animations are disabled
- Minimum feedback duration for UI interactions
- Selective cache invalidation for stale folder sizes
- Log Level dropdown styling to match other settings selects
- Browse cache invalidation, concurrency, and path handling improvements

## [2.1.0] - 2026-02-06

### Added
- **Recursive folder scanning** (#91) — Directory video counts now include files in subdirectories, so TV show folders display total episode count across all seasons
  - Non-blocking: Browse returns instantly using cached results, populates cache in background
  - Startup cache warming pre-computes all counts in a single efficient walk
  - Optimized with singleflight dedup, WalkDir, and semaphore (max 8 concurrent walks)
- **Runtime log level setting** (#92) — Change log verbosity (debug/info/warn/error) from the UI without restarting
  - New dropdown in Advanced Settings
  - `log_level` field on GET/PUT `/api/config` endpoints
  - Uses `slog.LevelVar` for atomic, concurrent-safe level switching

### Fixed
- **API batch job ordering** (#86) — Jobs submitted via API now queue in the order paths are submitted, not alphabetically
  - Replaced non-deterministic goroutine append + alphabetical sort with indexed pre-allocation
  - Each concurrent probe writes to its original position, preserving submission order
- **NVIDIA GPU setup simplified** — Pre-filled `NVIDIA_DRIVER_CAPABILITIES` and `NVIDIA_VISIBLE_DEVICES` env vars; users now just add `--runtime=nvidia`
- **Documentation overhaul** — Rewrote GPU setup guides, fixed stale API docs (browse response schema, encoder API, preset fields)

## [2.0.8] - 2026-02-05

### Fixed
- **4K HDR SmartShrink OOM crash** (#90) — VMAF analysis on 4K+ content no longer causes FFmpeg to be OOM-killed in containers
  - Content >1080p is now downscaled to 1080p before VMAF scoring, reducing memory from ~6GB to ~1.5GB per scorer
  - HDR downscale happens in linear light via combined zscale call (linearize + resize in one pass) for mathematically correct resampling
  - SDR downscale uses standard `scale` filter with aspect-ratio preservation (`-2:1080`)
  - Unknown/zero height defaults to 1080 as safety cap
  - Always uses HD VMAF model (`vmaf_v0.6.1`) since scoring happens at ≤1080p

### Changed
- Removed `SelectModel` function — replaced with `vmafModel` constant since only the HD model is ever used

## [2.0.7] - 2026-02-03

### Fixed
- **HDR SmartShrink support** — SmartShrink presets now work with HDR content regardless of `TonemapHDR` setting
  - VMAF scoring now tonemaps both reference and distorted legs to SDR for valid comparison
  - Previously only the reference leg was tonemapped, assuming distorted was already SDR
  - The `TonemapHDR` config setting only affects final transcode output, not VMAF analysis
  - HDR files with missing `color_transfer` metadata are still blocked (required for tonemapping)

### Changed
- Removed unused `tonemap` parameter from `BuildSampleEncodeArgs` — samples stay in native format during VMAF analysis

## [2.0.6] - 2026-02-03

### Changed
- **Parallel VMAF scoring** — Sample VMAF scores are now computed concurrently (up to 3 workers), significantly reducing scoring wall-clock time
- **Container-aware threading** — Thread allocation now uses `GOMAXPROCS` instead of `NumCPU`, correctly respecting container CPU limits
- **Removed CPU throttling** — VMAF scoring and sample extraction no longer run with `nice -n 19`; full CPU utilization improves iteration speed
- **Removed JSON stdout logging** — VMAF filter no longer outputs verbose JSON to stdout; scores parsed from FFmpeg stderr summary

## [2.0.5] - 2026-01-31

### Added
- **Multi-arch Docker builds** — Release images now include `linux/amd64` and `linux/arm64` architectures
- **Manual Docker build trigger** — Workflow dispatch option for on-demand builds
- **Browser refresh button** — Refresh current directory to detect file changes without full page reload

### Changed
- Documentation updated with FAQ for Mac users regarding Docker vs native builds and VideoToolbox availability
- Removed slow integration tests that required real encoding hardware
- Refactored dead code and consolidated DRY violations across job handling and config modules

### Fixed
- **Probe cache invalidation** — Cache now validates using inode+size to detect when files are replaced; refresh button clears cache and reprobes changed files
- **Refresh button animation** — Button performs smooth single rotation instead of barely-visible twitch

## [2.0.4] - 2026-01-30

### Fixed
- **MKV subtitle compatibility filtering** (#77)
  - Transcoding MP4/TS files with `mov_text` subtitles to MKV no longer fails
  - Subtitle streams are probed before transcoding; incompatible codecs are automatically filtered
  - Compatible codecs (srt, ass, ssa, subrip, webvtt, dvb_subtitle, hdmv_pgs) are preserved
  - Incompatible codecs (mov_text, eia_608, tx3g, ttml, dvb_teletext) are dropped with a warning log
  - Graceful fallback: if subtitle probing fails, transcoding proceeds with default behavior

## [2.0.3] - 2026-01-30

### Added
- **Same-codec re-encoding setting in UI** (#80)
  - New toggle in Transcoding settings to allow HEVC→HEVC or AV1→AV1 re-encoding
  - Useful for re-optimizing files encoded elsewhere with suboptimal settings
  - Previously only configurable via config file

### Changed
- SmartShrink presets now respect `allow_same_codec` setting
  - When disabled (default), SmartShrink skips files already in target codec
  - When enabled, SmartShrink will re-analyze and potentially re-encode for better compression

## [2.0.2] - 2026-01-30

### Added
- **Automatic encoder fallback chain** (#77)
  - If hardware encoder fails mid-job, automatically retries with next available encoder
  - Fallback order: Quick Sync → VAAPI → Software (varies by platform)
  - Per-job fallback—subsequent jobs still try preferred encoder first
  - Improves reliability when specific files trigger encoder issues

### Fixed
- VMAF concurrent analyses log no longer appears mid-splash screen
- Goroutine leak when Transcode errors occur before progress monitoring starts

## [2.0.1] - 2026-01-29

### Added
- **Configurable concurrent VMAF analyses** (#79)
  - New `max_concurrent_analyses` setting (1-3) in Advanced settings
  - Controls how many SmartShrink jobs can analyze simultaneously
  - Each analysis uses ~50% CPU (`numCPU/2` threads) for decode/filter operations
  - Multiple concurrent analyses stack (2 = 100% CPU, 3 = 150% over-subscribed)
- **Low-priority VMAF processing**
  - VMAF FFmpeg processes now run with `nice -n 19` (lowest priority)
  - SmartShrink analysis yields to other processes when system is busy
  - Regular transcoding still runs at normal priority

### Fixed
- VMAF analysis no longer pins CPU to 90%+ on idle systems
- Thread limiting via `-threads` and `-filter_threads` flags on all VMAF FFmpeg calls

## [2.0.0] - 2026-01-28

### Added
- **SmartShrink presets with VMAF-guided quality optimization** (#78)
  - New `smartshrink-hevc` and `smartshrink-av1` presets that analyze video content to find optimal compression
  - Uses VMAF (Video Multi-Method Assessment Fusion) to measure perceptual quality
  - Per-job quality selection: Acceptable (85), Good (90), Excellent (94) VMAF targets
  - Binary search algorithm finds the best CRF/bitrate that meets your quality threshold
  - Sample extraction at 5 positions (10%, 30%, 50%, 70%, 90%) with trimmed mean scoring
  - Supports all hardware encoders (NVENC, QSV, VAAPI, VideoToolbox) and software fallback
  - New job phase indicator: `analyzing` shown during VMAF analysis before encoding
  - SmartShrink results displayed in expandable job details after completion
- **New VMAF package** (`internal/ffmpeg/vmaf/`)
  - Automatic VMAF model detection with fallback support
  - HDR tonemapping support for VMAF analysis
  - Parallel VMAF analysis matching worker count
- **UI improvements**
  - Grouped preset dropdown (SmartShrink / Compress / Downscale categories)
  - Category prefix shown in closed dropdown (e.g., "Compress - HEVC")
  - Quality dropdown for SmartShrink presets
  - Preset and quality selections persist to localStorage
  - HW/SW badge in preset dropdown

### Changed
- Database schema updated to v6 (adds SmartShrink job fields)
- Presets now include `is_smart_shrink` and `skips_codec_check` flags
- Job model includes new fields: `Phase`, `SmartShrinkQuality`, `VMafScore`, `SelectedCRF`, `QualityMod`, `SamplesUsed`, `SearchIterations`

## [1.8.5] - 2026-01-22

### Added
- **Same-codec re-encoding option** (#76)
  - New `allow_same_codec` config option (default: false)
  - When enabled, allows HEVC→HEVC or AV1→AV1 transcoding for bitrate optimization
  - Useful for re-encoding at different quality settings without changing codec
  - Config-file only setting (not exposed in UI)

## [1.8.4] - 2026-01-18

### Added
- **Bit depth test file generation** - Added test vectors for 8-bit and 10-bit H.264/HEVC

## [1.8.3] - 2026-01-18

### Fixed
- **Queue sort options** - Minor UI polish

## [1.8.2] - 2026-01-16

### Fixed
- **Stuck decoder hang with unsupported codecs** (#75)
  - VAAPI hardware decode hangs indefinitely when codec isn't supported (e.g., AV1)
  - Added 10-second first-frame timeout watchdog that kills stuck FFmpeg processes
  - Allows existing software decode retry to trigger instead of hanging forever

## [1.8.1] - 2026-01-16

### Added
- **Queue sort options** (#74)
  - New unified filter/sort dropdown menu in queue header
  - Sort by: Default, File Name, Pre-Transcode Size, Post-Transcode Size, Space Saved
  - Click same sort option to toggle direction (↓ natural order, ↑ reverse)
  - Sort preference persists in localStorage
  - Pending jobs now show pre-transcode file size for sorting visibility

## [1.8.0] - 2026-01-15

### Added
- **HDR detection and tonemapping support** (#65)
  - Automatic HDR detection via color metadata (HDR10/PQ, HLG) with 10-bit BT.2020 fallback
  - Optional HDR-to-SDR tonemapping using software zscale filter (works on all systems)
  - HDR badge displayed on jobs and files in the UI
  - Tonemap toggle in Settings (default: OFF to preserve HDR metadata)
  - When tonemapping is disabled, HDR content preserves 10-bit p010 format and BT.2020 color
  - Hardware encoding still used when tonemapping - only the filter runs on CPU
- **Distinct "Skipped" status for already-encoded files** (#72)
  - Files already in target codec now show "Skipped" (yellow) instead of "Failed" (red)
  - Clearer distinction between actual failures and intentionally skipped files
  - Skipped jobs tracked separately in stats
  - "Clear by status" dropdown includes Skipped option

### Fixed
- **Data race in job event broadcasting** (#73)
  - Job events now use shallow copies to prevent race conditions
  - Fixes potential crashes when multiple subscribers receive events during job updates

### Changed
- Database schema updated to v4 (adds `is_hdr` column with automatic migration)
- SSE events now include HDR metadata for frontend display

## [1.7.5] - 2026-01-15

### Fixed
- **QSV/NVENC hardware acceleration broken in Docker** (#67)
  - QSV now tries direct init first, falls back to VAAPI-derived if needed
  - NVENC tries simple init first, falls back to explicit CUDA device init
  - Init mode detected once at startup instead of runtime retry logic
  - RTX 50 series can now attempt H.264 4:2:2 10-bit hardware decode

### Changed
- Simplified worker.go by removing HW init retry logic (detection handles it)
- Updated comments to clarify H.264 High10 (4:2:0) vs 4:2:2 10-bit support

## [1.7.3] - 2026-01-13

### Fixed
- **iOS Safari scrolling on iPad and iPhone** (#42)
  - Clear Queue and Transcode buttons no longer get pushed off-screen on iPad with many queue items
  - Fixed flex container chain with proper `min-height: 0` for scroll containment
  - iPhone landscape now shows both file list and action buttons (preset/transcode)
  - Uses `100dvh` for dynamic viewport height and `clamp()` for responsive list sizing

## [1.7.2] - 2026-01-13

### Fixed
- **FFmpeg -38 error at 97% progress** (#56)
  - H.264 10-bit content now automatically uses software decode (no GPU supports H.264 10-bit hardware decode)
  - Proactive codec detection skips hardware decode for unsupported profiles: VC-1/WMV3, MPEG-4 ASP
  - Adopted Jellyfin's battle-tested FFmpeg hardware acceleration pipeline
  - Removed problematic `vpp_qsv` filter that caused EOF errors with software decode
  - Hardware decode failures now automatically retry with software decode

### Changed
- QSV on Linux now derives from VAAPI for more reliable device initialization
- Consolidated duplicate `copyFile` implementations into shared `util.CopyFile()`
- Removed dead code (unreachable `vpp_qsv` scaling branch)

## [1.7.1] - 2026-01-11

### Fixed
- **Retry no longer leaves stale job entries in queue**
  - Added SSE "removed" event so all connected clients update when a job is deleted
  - Fixes duplicate/ghost jobs appearing after clicking Retry
- **Clear button now respects queue filter** (#55)
  - When a filter is active (e.g., "Failed"), Clear only removes jobs matching that filter
  - Clear button disabled when viewing Running filter (can't clear running jobs)

## [1.7.0] - 2026-01-11

### Added
- **Stop All / Resume button** (#49)
  - New button in queue footer to pause all running jobs
  - Running jobs are requeued and restart from the beginning when resumed
  - Auto-unpauses when new jobs are added to prevent accidental blocking
  - Button syncs state with queue (shows "Resume" when paused)

### Fixed
- **Queue scroll position preserved during job updates** (#51)
  - Progress updates no longer cause the queue list to jump to top
  - Incremental DOM updates instead of full rebuilds for smoother UI
  - Jobs correctly reposition when status changes (e.g., running → complete)
  - Status badge now updates from "Initializing" to "Running" when encoding starts

## [1.6.2] - 2026-01-09

### Fixed
- **iOS Safari bottom bar no longer clips UI**
  - Added `viewport-fit=cover` and `100dvh` viewport height for proper safe area handling
  - Fixes "Clear Queue" button being hidden behind Safari's redesigned bottom navigation in iOS 26

## [1.6.1] - 2026-01-09

### Fixed
- **Jobs no longer lost when reducing worker count** (#43)
  - When reducing concurrent workers (e.g., 3→2), running jobs are now requeued to the front of the pending queue instead of being lost
  - Added `Requeue()` method to safely return interrupted jobs to pending status

## [1.6.0] - 2026-01-09

### Added
- **MP4 output container option** (#44)
  - New setting in Advanced Settings to output MP4 instead of MKV
  - MP4 mode transcodes audio to AAC stereo and strips subtitles for web/direct play compatibility
  - MKV mode (default) preserves all streams as before
- Version displayed in startup splash screen
- Version field added to "Shrinkray started" log line for easier debugging

### Changed
- Updated CI to golangci-lint v2.8.0
- Refreshed splash screen text

## [1.5.0] - 2026-01-09

### Added
- **SQLite database for job persistence** - Jobs are now stored in SQLite instead of JSON for better performance and reliability
  - Automatic migration from queue.json to shrinkray.db on first run
  - JSON backup created before migration (queue.json.backup)
  - Supports large queues (1000+ jobs) with proper indexing
- **Session and lifetime space saved tracking** (#31)
  - Header shows session saved (resets when you choose)
  - Dropdown shows both session and lifetime totals
  - "Reset Session" button to start fresh count
  - Lifetime total persists forever across restarts and queue clears
- Schema versioning for future database migrations

### Changed
- Job queue now uses SQLite with WAL mode for better concurrent access
- Stats calculation moved from in-memory to database queries
- Updated golangci-lint config for v2 format

### Fixed
- Running jobs correctly reset to pending on restart (#35)

## [1.4.12] - 2026-01-08

### Fixed
- Fix running jobs skipped on container restart (#35)
  - When container restarts while jobs are encoding, those jobs now correctly reset to pending
  - Reset state is immediately persisted to disk so it survives further restarts
  - Jobs restart from the beginning in the correct order

## [1.4.11] - 2026-01-08

### Fixed
- Fix QSV encoding failure on MKVs with embedded cover art (#40)
  - MKV files with attached pictures (cover art) have multiple video streams
  - Previous `-map 0` copied all streams, causing QSV to fail encoding JPEG images
  - Now uses explicit stream mapping: `-map 0:v:0 -map 0:a? -map 0:s?`
  - Only maps primary video, all audio, and all subtitles (skips cover art)

### Changed
- Code cleanup: consolidated duplicate utilities, removed dead code, documented magic numbers

## [1.4.10] - 2026-01-08

### Fixed
- All hardware encoders (QSV, VAAPI, NVENC) now automatically retry with software decode when hardware decode fails (#38, #32)
  - Fixes AV1 files on pre-11th gen Intel which can't hardware decode AV1
  - Fixes VC1, MPEG4-ASP, and other codecs that hardware decoders don't support
  - First attempt uses full hardware acceleration
  - If decode fails, automatically retries with software decode + hardware encode
  - Users with capable hardware still get full HW acceleration
  - Fallback is transparent—no user intervention required

## [1.4.9] - 2026-01-08

### Fixed
- Fix QSV encoding failures when hardware decode falls back to software (#32, #38)
  - Changed filter chain to `format=nv12|qsv,hwupload=extra_hw_frames=64`
  - The pipe syntax accepts either CPU frames (nv12) or GPU frames (qsv) without forced conversion
  - HW decode path: zero-copy passthrough (no performance impact)
  - SW decode path: frames uploaded to GPU via hwupload
  - Matches the pattern used by Jellyfin and our working VAAPI implementation

## [1.4.8] - 2026-01-08

### Fixed
- QSV encoding now properly handles software decode fallback (#32, #38)
  - Previous fix using `vpp_qsv` was incorrect - it can't accept CPU frames
  - Now uses `hwupload=extra_hw_frames=64` like Jellyfin does
  - Added `-init_hw_device qsv=qsv -filter_hw_device qsv` for proper device initialization
  - VC1 and other codecs that QSV can't hardware decode now transcode correctly
  - Note: This fix was incomplete - see 1.4.9

## [1.4.7] - 2026-01-08

### Fixed
- QSV encoding now works for compress presets when hardware decode falls back to software (#38)
  - Added `vpp_qsv` base filter to handle CPU-to-QSV frame upload (note: this fix was incomplete, see 1.4.8)
  - Previously only worked for downscale presets (1080p/720p)

## [1.4.6] - 2026-01-07

### Fixed
- Speed and ETA now display correctly during VAAPI hardware encodes (#29)
  - When FFmpeg reports N/A for time/speed, calculate from frame count and elapsed time

## [1.4.5] - 2026-01-07

### Fixed
- Progress bar now works during VAAPI hardware encodes (#29)
  - FFmpeg reports `N/A` for time-based stats with some hardware encoders
  - Added frame-based progress calculation as fallback when time is unavailable

## [1.4.4] - 2026-01-07

### Added
- `keep_larger_files` config option to keep transcoded files even if larger than original
  - Useful for users who want codec consistency across their library

## [1.4.3] - 2026-01-07

### Added
- Preserve original file modification time after transcoding (#33)
  - Fixes compatibility with Unraid Mover Tuning and similar tools that use mtime for file age

## [1.4.2] - 2026-01-07

### Fixed
- QSV scaling now works on Intel UHD 630 and similar iGPUs (#21)
- Replaced `scale_qsv` with `vpp_qsv` using explicit dimensions for downscale presets

## [1.4.1] - 2026-01-06

### Fixed
- QSV (Intel Quick Sync) encoding now works when hardware decode falls back to software (#21)
- VAAPI encoding more reliable with mixed hardware/software decode scenarios (#21)

## [1.4.0] - 2026-01-06

### Added
- Structured logging system with configurable log levels (debug, info, warn, error)
- `log_level` config option to control logging verbosity
- Job lifecycle logging (started, completed, failed, cancelled)
- FFmpeg command logging at debug level for troubleshooting
- FFmpeg stderr capture for better error diagnostics

### Fixed
- Duplicate "Job started" log entries when running multiple workers

## [1.3.7] - 2026-01-06

### Fixed
- "Processing files..." banner no longer persists after single-file jobs start

## [1.3.6] - 2026-01-05

### Fixed
- VAAPI encoding on AMD GPUs now works when hardware decoding falls back to software (#21)
- Added `hwupload` filter to VAAPI transcode pipeline for proper frame format handling

## [1.3.5] - 2026-01-05

### Fixed
- VAAPI hardware encoder detection now works on AMD GPUs (#21)
- Test encode now properly uploads frames to GPU memory before encoding

## [1.3.4] - 2026-01-05

### Fixed
- Auto-detect `/temp` mount in Docker - no manual config needed (#23)
- Downscale presets (1080p/720p) no longer incorrectly skip files already in HEVC/AV1

### Added
- `TEMP_PATH` environment variable support for explicit temp directory override

## [1.3.3] - 2026-01-05

### Fixed
- Auto-create config file on first run with correct absolute paths
- Default `queue_file` now uses absolute path `/config/queue.json` to prevent permission errors

## [1.3.2] - 2026-01-04

### Fixed
- Breadcrumb navigation now shows full path instead of always collapsing to "..." (#22)
- Users can navigate to any intermediate folder, not just Home or immediate parent
- Deep paths (>3 levels) show last 3 folders: `Home / ... / Parent / GrandParent / Current`

## [1.3.1] - 2026-01-04

### Fixed
- Quality slider labels were backwards - now correctly show "Higher quality" on left and "Smaller file" on right (#24)

## [1.3.0] - 2026-01-01

### Added
- Scheduling feature to restrict transcoding to specific hours (e.g., overnight only)
- Schedule settings in Settings panel with enable toggle and time selection
- Schedule status display near transcode button when enabled
- Advanced settings section for quality control (CRF values for HEVC and AV1)
- Collapsible advanced settings panel (closed by default)

### Fixed
- Quality input fields now validate on change instead of every keystroke

## [1.2.0] - 2025-12-29

### Added
- Dark mode toggle
- Infinite scroll for queue display (handles thousands of jobs smoothly)
- Real-time progress feedback for batch file processing
- Allow selecting folders with nested videos for batch transcoding
- Queue sorted by status (running jobs first, then pending, then completed)

### Fixed
- UI no longer hangs when adding thousands of files to queue

## [1.1.0] - 2025-12-28

### Added
- Skip files already encoded in target codec (HEVC/AV1) to prevent unnecessary transcoding
- Skip files already at target resolution when using downscale presets (1080p/720p)
- Version number displayed in Settings panel

## [1.0.0] - 2025-12-25

### Added
- Initial public release
- Hardware-accelerated transcoding (VideoToolbox, NVENC, QSV, VAAPI)
- HEVC and AV1 compression presets
- 1080p and 720p downscale presets
- Batch folder selection for entire TV series
- Async job creation to prevent UI freezes
- Pushover notifications when queue completes
- Retry button for failed jobs
- Mobile-responsive stats bar
- Queue persistence across restarts