/exp-status

To monitor running experiments, the system scans wiki experiment pages and checks process liveness or log tails for local screen sessions and remote SSH jobs, classifying each run as running, anomalous, or pending collection.

Anomaly detection for research pipelines works by scanning experiment log tails to flag patterns like NaN, OOM, Traceback, or Inf, reporting these issues without auto-fixing so users can manually decide the next action.

Yes, you can schedule automated experiment monitoring as periodic cron checks by scoping the status checks to a specific research pipeline, ensuring continuous tracking without manual intervention.

To auto-advance a research pipeline, the monitor checks if all experiments within the specified pipeline have completed, then automatically triggers the pipeline advancement process to proceed to the next stage.

The best way to collect results from completed screen runs is using the collection-ready monitor flag, which identifies finished experiments pending collection and triggers the result collection process automatically.

Experiment monitoring supports both local screen-based runs and remote GPU SSH jobs, checking process liveness and log tails across both environments to provide a unified status overview.