Run and lifecycle
--wait polls until a terminal status. An omitted evaluation ID resolves to the
most recent workspace evaluation. cancel applies only to queued or running
evaluations.
eval run also forwards an optional [eval.feedback_capabilities] table from
the selected manifest. Capability names and values are competition-defined;
unsupported dataset or race combinations fail validation.
Results and diagnostics
Artifact downloads
all; if omitted, all available
artifacts are selected. The default destination is inside the selected
experiment. See Evaluations for examples and redaction
behavior.