Performance
- Interactive filtering now uses cached transcripts only; falls back to raw session fields without generating new transcripts.
- Demoted heavy background work (filtering, indexing, parsing, search orchestration) to
.utilitypriority for better cooperativeness. - Throttled indexing and search progress updates (~10 Hz) and batched large search results to reduce main-thread churn.
- Gated transcript pre-warm during typing bursts, increased interactive filter debounce, and debounced deep search starts when typing rapidly.
- Built large transcripts off the main thread when not cached, applying results on the main thread to avoid beachballs.
Documentation
- Added
docs/Energy-and-Performance.mdsummarizing performance improvements, current energy behavior, and future options.