OPERATOR GUIDE
Troubleshooting
Start with structured status and source availability; do not weaken safety to make a warning disappear.
Configuration rejected
Check required keys, YAML types, unknown fields, path conflicts, existing SOP directory and mandatory dry_run: true. Never quote booleans or integers to bypass type checks.
Collector PARTIAL or UNAVAILABLE
- inspect the ReasonCode and bounded error
- confirm the kernel source exists
- confirm driver or firmware exposure
- check runtime read permissions
- install optional utilities only through normal change control
- accept legitimate absence rather than fabricating telemetry
SMART and NVIDIA
SMART requires usable smartctl JSON and device access. NVIDIA proprietary metrics require a working driver and nvidia-smi; hardware under nouveau should appear as detected with NVIDIA_PARTIAL.
Audit mismatch
Stop promotion, preserve both backends, inspect the latest safe errors and storage availability, and avoid claiming a fully audited cycle until parity is restored.