Validate Ticket Output

Validate Ticket Output

A support tool asks a language model to classify each ticket and copy its order ID, and the model replies with JSON text. Before the application stores or routes anything, it must check that reply. Write the check.

Implement validateTicket(raw, source). raw is the model's reply as a string, and source is the original ticket text. Return:

• { ok: true, value: { category, order_id } } when the reply is accepted
• { ok: false, kind, reason } for the first failing layer, where kind is "parse", "schema", or "source" and reason is a short, non-empty message of your choice

The contract

Check the layers in this order and stop at the first failure:

ID policy. Find the order IDs in source: ORD- followed by exactly three digits, as a whole word (so ORD-1234 contains no valid ID). Count distinct IDs:

• none: order_id must be null
• exactly one (repeats count once): order_id must be that ID
• more than one: order_id must be null and category must be "other"

Fixtures

The Expected column is what the validator returns. Whether each category is the right one for its ticket is a separate question.

Notes

• An empty string is not a substitute for null: "order_id": "" is a schema failure.
• A reply wrapped in a Markdown code fence is not valid JSON.

In Python, implement validate_ticket(raw, source) and return dicts with the same keys ("ok", "value", "kind", "reason").