Status

The live operational state of the block is not kept in these docs — it lives in the files the scripts read and write, so it never goes stale.

Source of truth

config.yaml is one-shot per run — every key is configuration, and it is not edited during execution. Live state lives elsewhere:

• artifacts/index.yaml — completed archive records with status completed | failed | interrupted. Entries are appended automatically by scripts/archive_run.sh when a run exits; this file does not represent an in-progress run.
• artifacts/jobs/<job>/ — the in-progress and finished per-task outputs of the active job.
• dashboard — a live visual view of jobs, analysis reports, and trajectories.

Read the newest entry of artifacts/index.yaml for the last archived run, and artifacts/jobs/<job>/ (plus the dashboard) for what is happening now.

During a run

While a job is active, monitor artifacts/jobs/<job>/ and the dashboard. Do not edit config.yaml mid-run — it is configuration only. The terminal status is appended to artifacts/index.yaml when the run exits.

Remote execution

eval runs on the node declared in config.yaml → meta_info.resources.ip (the checked-in default is local).

• For local / null, run scripts directly in a named tmux session (e.g. tmux new-session -d -s eval …).
• For a configured remote hostname/IP, connect using SSH keys or ~/.ssh/config and operate inside a tmux session there. Keep host details and credentials in a private run profile.

Either way, all execution must happen on the configured IP in a named tmux session so the run survives shell disconnects. Environments and the Harbor checkout are host-local — never build them on a host the run won't use.