Reliability

Reliability is more than adding a try/catch. A caller must distinguish an immediate result, submission rejection, task failure, wait timeout, and long-term trend, with a suitable observation path for each.

1. Failure observability: combine futures, failure callbacks, cumulative counts, and recent events.
2. Monitoring and sampling: collect task statistics or a full lifecycle snapshot with the required precision and acceptable overhead.
3. Troubleshoot by symptom: use a fixed check order for no execution, backlog, wait timeout, stalled shutdown, real-time drops, and unavailable GPU.
4. Linux, Windows, and Android deployment: verify builds, CPU availability, real-time permissions, memory locking, Android CPU-only fallback, and platform-specific behavior actually take effect.

Communication drops, overwrites, stale reads, and missed phases remain component-local events; they do not automatically enter ExecutorFailureStatus. Define their observation path in choose a communication component.