What should a lab put in a request for a review sample?

Use this checklist to turn a broad interest in business reviews into a concrete sample request.

- State the target model behaviour. - Name the business domain and reviewer role. - Describe the agent's available context window. - Specify document-reading and tool capabilities. - Identify acceptable human intervention points. - Describe the baseline used for comparison. - State the evidence needed for a proceed decision.

The operating conditions affect what a sample can demonstrate. A model receiving a complete evidence bundle faces a different task from an agent that must discover records through tools. A sample intended for supervised training also needs a different presentation from a hidden-answer evaluation episode.

Include one failure the lab wants to understand, such as losing an obligation after a handover or drawing a conclusion from insufficient evidence. That gives the sample a specific learning objective without prescribing its entire construction.

A buyer can ask MirrorFirm to respond with a proposed generated sample design and any assumptions that need agreement. The first inspection can then compare the design with the intended experiment before the lab decides whether to broaden the scope.