Budgeted Two-Stage Pre-Validation for Retrieval-Grounded Machine Learning Pipelines: A Group-Held-Out Evaluation
Retrieval-grounded machine-learning pipeline synthesis must reconcile code templates with package versions and resource budgets. We study a two-stage pre-validator adapted from concrete dependency-satisfiability and resource-prediction mechanisms: an inexpensive retrieval-profile margin ranks candidate manifests; a bud...