Skip to content
MirrorFirm

What should a pilot brief for generated allocation material contain?

A checklist of the items a buyer can put in a short brief, from decisions to test and file formats to intended uses, tool signatures and the review process for disclosure risk.

A short brief makes a pilot faster to scope and easier to judge; include the following in yours.

- The allocation decisions you want to test, such as short payments, overpayments and reversals. - The domain framing and any vocabulary your agents must handle. - Delivery form: dataset, task pack, simulated environment, or a combination. - File formats and schema constraints, such as JSONL for tasks and Parquet for snapshots. - Tool actions your agent framework expects, with argument names. - The reference outcomes and rationale you need, and the scoring you plan to apply. - The uses you intend, including training, internal evaluation or publishing results, so licence terms can be discussed. - Your review process for grounding and disclosure risk, and who signs off.

Send the brief with two or three example tasks you consider typical and one you consider hard. This gives the provider something concrete to match, and gives you a basis for judging whether the sample reflects your intent. Agree what counts as a pass for the pilot before it starts, so the decision at the end rests on criteria you set in advance rather than on impressions formed during the review.

A concrete brief with typical and hard example tasks makes the pilot faster to scope and easier to judge.

Audience:
Evaluation teams
Content type:
Checklist
Buyer question:
Scope and getting started

Content release: r-a220eb6e099b243b

More on Payment allocation