An engineer needs to process unstructured source files through Spark without pretending that every file is a relational table. Which resource should be considered?

1. A volume
2. A model deployment endpoint
3. A single scalar notebook parameter
4. An agent's final response