A user who discovered an overnight AI job had stalled prompted suggestions to separate external monitoring from human approval. An internal self-check can stop with the job, so a separate scheduler should detect missing completion signals and delays, while steps such as login are handed to a person. Healthchecks documentation describes alerting when a scheduled success signal fails to arrive. Process-alive signals should be distinguished from verified output completion, and a commenter’s personal implementation is not a security validation.