Load a single ML Training Pipeline by id, fully typed for mutation-free
reads. Returns null when not found.
pipeline primary-key value
OptionalcontextUser: UserInforequest user
Optionalprovider: IMetadataProvideroptional provider for multi-provider correctness
Compute the next monotonic model version under a pipeline — max(Version)+1,
or 1 when the pipeline has no prior models. Implemented as a narrow,
read-only count/scan over MJ: ML Models.
pipeline primary-key value
OptionalcontextUser: UserInforequest user
Optionalprovider: IMetadataProvideroptional provider for multi-provider correctness
OptionalresolveResolve the sidecar driver key for an algorithm from its
MJ: ML Algorithms.DriverClass (e.g. xgboost, logistic_regression).
The pipeline view exposes Algorithm as the algorithm's display Name
(XGBoost, Multilayer Perceptron), which is NOT what the sidecar registry
keys on — it requires the lowercase DriverClass. The engine therefore
resolves the driver key by id rather than trusting the display name.
Optional so in-memory test fakes (which set pipeline.Algorithm directly to
the driver key) need not implement it — the engine falls back to
pipeline.Algorithm when this is absent or returns null.
MJ: ML Algorithms primary-key value
OptionalcontextUser: UserInforequest user
Optionalprovider: IMetadataProvideroptional provider for multi-provider correctness
the DriverClass driver key, or null when not found
Read seam for loading existing records (the training pipeline definition and the next-version lookup). Wraps
RunViewin production; tests inject canned rows. Never throws for logical read failures (mirrorsRunView).