Choose a source and target

Start with the business boundary: where the input already exists, how many rows may be read, and which side effect is required. Then choose the matching typed Authoring source and target; do not start from an exporter name.

Decision path

1. Use generated when no input dataset exists and the requested cardinality is fixed.
2. Use a file source for project-relative data and memstore for an earlier product in the same run.
3. Use database, mongodb, kafka, rabbitmq, or object_storage only when that configured client owns the input.
4. Set count for an exact bounded read or max_count for a hard upper bound.
5. Give iterate no target. When traversal must emit a product, put a target-bearing generate in its child context.

Complete workflows

• Project file to artifact
• Relational copy
• MongoDB projection
• Kafka forwarding
• RabbitMQ forwarding
• Object-storage transformation

Client IDs in Authoring refer to clients declared by the deployment or surrounding setup. Connection details, credentials, broker topology, and buckets are not copied into the business intent.

Review checklist

• The source family matches the actual ownership boundary.
• The read is bounded.
• The selected distribution is supported by that source family.
• Every target represents an intended side effect.
• Source and target clients remain separate when permissions or systems differ.