ct run monitor
Score an eval-backed source's recorded transcripts with monitors
Usage
ct run monitor [OPTIONS] TRAJ_LIKE
Arguments
| Argument | Type | Required |
|---|---|---|
TRAJ_LIKE | text | Yes |
Options
| Option | Description |
|---|---|
-m, --monitor TEXT | Monitor selection (repeatable): a name, a .md asset path, or a module:function import path. Prefix with action/ or trajectory/ when the name alone is ambiguous. ct protocols monitors to see available. [default: mean-action, mean-quick] |
--trusted-model TEXT | Model bound as the eval-level trusted role the monitors run on. [required] |
-o, --save-dir PATH | Directory to save monitored trajectories (default: ./data/monitoring/{monitor}/{traj_name}_{timestamp}) |
--upload / --no-upload | Upload results to the Control Tower database (enabled by default). [default: upload] |
--run-name TEXT | Name for the run (default: monitor-{monitor_name}) |
--tag TEXT | Tags for the run (can be specified multiple times) |
-L, --limit INTEGER RANGE | Limit the number of trajectories to monitor [x>=1] |
-C, --max-concurrent INTEGER | Max concurrent monitor calls (default: 16) |
--overwrite | Replace a selected monitor's verdicts when the source already carries them; without it that source is refused. |
--cot-level [none|reasoning|extended] | Chain-of-thought visibility for monitors: none, reasoning (models output text), or extended (reasoning and CoT). [default: none] |
--help | Show this message and exit. |