A small example.
A useful next step.
Start with an inspectable source and an expected outcome. Adapt the workflow in your own implementation rather than treating a successful sample as a complete integration.
Customer record mapping
Rename columns, trim whitespace, preserve string identity, and report one missing required value. The working quickstart runs this fixed fixture locally.
Explore this workflowDocument-to-knowledge review
Choose a manual, preserve headings and source identity, and define questions with expected source sections before splitting content into retrieval units.
Explore this workflowPrompt registry handoff
Keep the definition separate from actual executions. Preserve roles, variables, a version, and review status before promoting an application instruction.
Explore this workflowSmall enough to understand at a glance.
Synthetic source records, explicit output, and a prompt record you can inspect.
Customer source · JSON
Three original records before mapping. One required identifier is deliberately missing.
Download example ↗Customer source · CSV
The same source values in a header-based CSV. Use a proper CSV parser in your own implementation.
Download example ↗Validation result · JSON
The exact accepted and rejected output from the working quickstart.
Download example ↗Customer contract · JSON Schema
The example record shape, nonempty string checks, and three allowed country labels.
Download example ↗Prompt definition · JSON
A synthetic support-summary definition with messages, declared variables, and a version.
Download example ↗Keep an expected result beside every fixture
A useful example includes its failure behavior. When expanding these fixtures, write down whether a new case should be accepted, rejected, or require review. Test missing values separately from explicit empty values and test identifiers that must retain leading zeroes.
For document and prompt workflows, an acceptable result involves meaning as well as shape. Use reviewed source sections and task-specific evaluation cases. No document extraction or model inference is performed by these downloads.