What should a buyer request in a sample package?
A reviewable sample package should expose the task contract, expected outputs and known limits, giving technical reviewers enough context to decide whether a pilot is workable.
Ask for a sample package that lets a technical reviewer complete and inspect the proposed task end to end.
- A plain description of the target model behaviour. - Generated input records with their supporting context. - A task prompt and required response structure. - An annotated reference response with reasoning. - An intentionally incomplete case and expected handling. - A statement of known exclusions and simplifications. - A manifest identifying the package's constituent files.
The reference response should show what a successful submission looks like, including how evidence is cited and how unresolved points are expressed. It should not be merely a polished narrative that leaves the evaluator to infer the actual acceptance criteria.
Known simplifications matter because a sample may omit parts of the wider business process. A logistics review task might begin after delivery reconciliation and therefore say nothing about how discrepancies were originally detected. That boundary should be visible during acceptance.
The buyer's first review can check whether the package is understandable without private explanations from its author. Any required clarification should feed back into the task contract so later users and automated evaluation systems receive the same essential context.
Request a sample that supports an acceptance decision, including its limitations and an inspectable reference response.
- Audience:
- Model labs
- Content type:
- Checklist
- Buyer question:
- Scope and getting started
Content release: r-a220eb6e099b243b
Related questions
- What integration contract should we establish before loading a pilot release?
- What makes a useful first pilot scope for an agent team?
- What licence questions should we resolve before using generated evidence material?
- How can a pilot reveal that the agent needs a different action boundary?
- What should a sample request include before a pilot is agreed?
- What integration contract should be agreed before a pilot build?