Can a task pack include customer requests that are missing essential details?
A generated task pack can include incomplete requests, answerable clarification paths and explicit limits on what the model may infer.
**Can the customer message be incomplete?**
Yes. A task pack could contain a request such as changing the delivery location for an order without specifying whether the change applies to the current shipment or future orders. The ambiguity becomes useful when those choices lead to different actions.
**What else should the pack contain?**
The generated state can include an account preference, an open order and a shipment already in preparation. A response branch can supply the missing scope when the model asks a relevant question. A task author can define which facts are known, which remain unknown and which may be obtained through available interactions.
**How does this differ from withholding arbitrary information?**
The task should make a productive next step possible. If the only valid action requires a fact that no tool or customer response can provide, the pack should explicitly support a blocked outcome.
MirrorFirm could generate alternative customer replies that resolve the same initial ambiguity in different ways. This allows training on concise clarification: asking for the missing decision variable, avoiding repeated questions about facts already present and proceeding once enough information is available.
Teach targeted clarification by making missing information consequential and recoverable through the task.
- Audience:
- Model labs
- Content type:
- FAQ
- Buyer question:
- What can be generated
Content release: r-a220eb6e099b243b
Related questions
- What evidence artefacts should accompany a generated change request?
- Can generated environment state include disagreement between connected systems?
- Which relationships can a generated dataset contain around a customer change?
- How can generated change material be packaged for an agent harness?
- How can a task pack represent a change governed by different policy versions?
- What policies and authority relationships could a generated environment contain?