{
  "context": "This criteria evaluates how well an engineer uses Kedro's pipeline and data catalog APIs to implement automatic failure analysis and resume point determination. It focuses on correctly identifying which datasets persist across pipeline failures and using this information to determine optimal resume points.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Pipeline node access",
      "description": "Uses Pipeline.nodes or Pipeline.node_dependencies to access the nodes in the pipeline and understand the pipeline structure",
      "max_score": 15
    },
    {
      "name": "Dataset persistence detection",
      "description": "Correctly identifies whether datasets persist to disk by checking dataset types in the DataCatalog (distinguishing MemoryDataset from persistent datasets like CSVDataset, ParquetDataset, JSONDataset, etc.)",
      "max_score": 25
    },
    {
      "name": "Node dependency traversal",
      "description": "Uses node.inputs and node.outputs attributes to traverse the pipeline dependency graph and identify upstream dependencies of the failed node",
      "max_score": 20
    },
    {
      "name": "Catalog dataset lookup",
      "description": "Uses DataCatalog methods (e.g., catalog._datasets or catalog.list()) to access dataset definitions and check their types",
      "max_score": 15
    },
    {
      "name": "Resume point logic",
      "description": "Implements correct logic to determine resume points: finds the furthest downstream nodes whose outputs are not persisted (or have no persisted outputs), ensuring all required upstream persisted data is available",
      "max_score": 20
    },
    {
      "name": "Error handling",
      "description": "Validates that the failed_node_name exists in the pipeline and raises ValueError if not found, as specified in the API",
      "max_score": 5
    }
  ]
}