Class RunFixtures

For every harness bound since the last cleanup: look up the pipelines of its agent runs again by their exact names (a run can create them after the client call returned), delete the pipelines and models it recorded with their training runs and scoring bindings through the entity layer, put back the File Storage Provider it switched on, and close its pool. Every step is attempted; throws one error naming each failure. No-op when nothing was bound.

The run's harness on a provider the caller already holds (an mj test check's client provider), built on first use and shared while the provider stays the same. A different provider gets its own harness, and the fixture models are forgotten: they were recorded on the previous harness, which is still cleaned up by CleanupRun.

Trains a logistic-regression model on the given dataset (MaxRows: 1_000_000) and returns its ModelID.

Memoised W1 s42 Draft logistic-regression model (MaxRows: 1_000_000).

Memoised W1 s42 Validated logistic-regression model (MaxRows: 1_000_000), promoted via h.Promote.