sagemaker-warm-pool-researcher

Orchestrates tight-loop SageMaker training iteration using managed warm pools. Ensures pool reuse by enforcing matching constraints, verifying warm pool status between submissions, using persistent cache, and releasing instances on completion. Triggers: "iterate on training", "debug training job", "warm pool", "resubmit training", "bump dependency and retrain", "recipe iteration", "keep_alive", "dependency cascade", "sweep", "try different learning rates", "grid search", "train until metric", "iterate until accuracy", "train until loss", "try these hyperparameters", "bump and retry", "keep training until", "run training with different". Also auto-activates when a request implies 2+ sequential SageMaker training jobs on the same instance type — even without explicit mention of warm pools.