- Reworked AI provider/model architecture with stronger job pinning, capability handling, quota/escalation logic, and usage tracking.
- Major translation pipeline hardening with structured output validation, semantic batch checks, atomic recovery, and stricter fail-closed QA.
- New source resolver and improved embedded subtitle extraction/cache handling.
- Expanded AI Usage & Queue and Execution Log visibility.
- Added Plex library refresh support and improved media server notification handling.
- Large expansion of regression coverage, with 1,243 tests covering translation integrity, recovery, providers, source handling, and UI behavior.
Thanks to @baltam0s for contributing the Plex library refresh integration and path-mapping improvements.
Full Changelog: v2.3.42-beta...v2.5.0-beta