# Korra illustrative record pack

Version 1.0. Created 6 October 2026.

These three records are invented by Korra. They contain no real customer records, private messages, employee feedback, or business exports. Synthetic means made up. These examples do not come from transformations of real customer data.

## Files

- `records.jsonl`: three task records. JSONL means one JSON object on each line.
- `schema.json`: a JSON Schema that defines the example fields and required types.
- `README.md`: this format guide.

## What each record contains

- `schema_version`: the version of this example structure.
- `synthetic`: always true in this pack.
- `record_id`: a unique example identifier.
- `record_type`: support, operational workflow, or AI review/feedback.
- `provenance`: the invented source, author, date, and explicit real-data flag.
- `task_context`: the role, rules, and facts that explain the task.
- `request`: what needs to be done.
- `work`: the actions taken in this invented case.
- `result`: the example status and outcome.
- `evaluation`: checks for an acceptable result and allowed alternatives.
- `feedback`: present in the AI review example; links the draft, revision, decision, and reason.
- `quality_notes`: limits that stay with the example.

## Using the examples

You may use these made-up examples to inspect or test the format. They do not grant permission over any business dataset. The pack is not a dataset offer, customer case study, available inventory, or measured AI result.

For an evaluation demonstration, provide only `request` and relevant `task_context` to the system being checked. Keep `work`, `result`, `evaluation`, and `feedback` outside its input when they expose the answer. The checks here are illustrative; they have not been validated as a benchmark.

Training rights, real source history, selection rules, privacy controls, quality checks, and delivery requirements must be agreed separately for a real dataset. A record structure alone does not establish those rights or checks.

Learn more at https://korra.io/use-cases and https://korra.io/dataset-quality.