The algorithm NAME the chosen experiment proposes (resolved to AlgorithmID by the builder).
Point-in-time assembly strategy.
Plain-language description carried onto the pipeline row.
The assembled FeatureStep DAG.
Leakage protection (deny-list + dominance threshold).
Human-readable pipeline name (derived from the plan's Goal).
Classification or regression.
Source bindings the features draw from.
The training-unit entity NAME (resolved to TargetEntityID by the builder).
Label column / expression.
Validation strategy.
The resolved, ready-to-persist configuration for one
MJ: ML Training Pipelinesrow.