The image subtitle walker now uses dedicated range-prefetch endpoints that return all subtitle frames overlapping a requested time window, allowing the client to cache multiple caption images from one backend call.\n\nAuthenticated and shared video playback now use adaptive range windows tuned for the new model: smaller near playback and larger when already far ahead. Route, client, and parser tests were extended to cover the new range APIs and prefetch path.
Image subtitle prefetching now supports adaptive step sizing and adaptive pacing so the walker uses smaller strides near playback and larger strides when cache lead is high.\n\nApp and share playback are configured to run this adaptive mode (1s near playback up to 5s far ahead) to improve cache fill throughput while reducing backend request pressure. Shared-player tests include regression coverage that verifies stride growth in adaptive mode.
The image subtitle prefetch walker now supports a configurable stride so clients can reduce subtitle frame prefetch request volume.\n\nApp and share playback now use a 2-second prefetch step for cache-walker requests, which lowers backend call frequency while keeping normal subtitle readability paths intact. A regression test confirms configured larger steps are respected, and technical references were updated accordingly.
Video preview initialisation now uses a lifecycle token so stale async setup from a previously closed preview is discarded and cleaned up immediately. This prevents old preview instances from leaving behind active client walkers or stream sessions when users switch videos quickly.\n\nImage subtitle prefetching now also extends to the known video end as soon as duration metadata arrives, even before playback advances, so subtitle cache filling continues towards full coverage without relying on current playback position.
Subtitles no longer lose their lookahead buffer when cache-only fallback fetches are triggered repeatedly during playback. This improves subtitle continuity in longer viewing sessions and when moving between videos, where captions could otherwise appear to stall until playback moved far enough ahead again.\n\nThe shared subtitle player now preserves prefetch progress during fallback current-frame requests, and a regression test covers repeated fallback triggers to ensure prefetch requests are not reset by that path.
Image subtitle rendering now tolerates timestamp boundary drift when matching cached frames and can optionally perform a throttled current-time fallback fetch when cache-only mode misses. This improves subtitle reliability near the end of videos or while paused without introducing request bursts. Added regression coverage for paused near-end rendering and cache-miss fallback behavior.
The Debug dashboard now shows server-maintenance telemetry, including auth/share row counts and restart scheduler status. The Admin server section now includes a persistent scheduled-restart form with cadence/time options, and controls are disabled when the app is not running under systemd.
Admins can now persist a global scheduled restart policy with cadence and time. When enabled under systemd, the server waits for at least one hour of no logged-in clients before shutting down so systemd can restart it safely. The debug API now exposes maintenance row counts and cleanup/scheduler runtime snapshots for long-uptime monitoring.
Server maintenance now removes stale auth sessions and expired share links automatically during long runtimes. The app also prunes stale zip cache files in the background so disk usage and DB row growth stay under control over weeks of uptime.
Server-side image subtitle caches now evict frame and event data as soon as the slowest active viewer has passed each event (with a short grace), instead of keeping old event bytes until cache-cap pressure.\n\nThis reduces long-session memory growth and lowers crash risk when subtitle prefetch advances quickly through many events.
Image subtitle warmup now starts immediately when a subtitle track is selected, prefetching an initial timeline window before playback advances. This reduces missing caption frames at the beginning of playback for longer videos and avoids needing to pause or rewind to catch early lines.\n\nThe app and share player both use a 60-second startup warmup while keeping cache-only foreground subtitle rendering and background walker ownership of subtitle network requests.