For buyers
Start with one consequential workflow.
A useful pilot begins with a specific decision: what reliability question must the evaluation answer? Maydan then scopes the workflow, evidence and boundaries with the buyer.
A pilot scoping conversation
- Name the workflow and decision. Which government or HR process matters, and what would the evidence change for your team?
- Define the agent interface. What application interaction and observation format should the agent use?
- Agree on pass conditions. Specify the terminal state, necessary prerequisites, audit requirements and records that must remain untouched.
- Set the evidence boundary. Identify rule sources, synthetic assumptions, excluded cases, language, task count and the traces or verifier outputs to review.
- Review feasibility. Confirm the proposed environment work, delivery plan and commercial terms before any commitment.
What may be delivered?
Depending on the agreed scope: resettable synthetic instances, task definitions, deterministic verifiers and a reviewable evidence package. Exact work and timing are defined per pilot; the public sampler is a separate, narrower artifact.
Questions worth asking us
- How is a “do nothing” case represented and scored?
- How are bystander records protected in the verifier?
- Which modeled rules are official, inferred or synthetic?
- What does an unsuccessful agent trajectory reveal?
- Which parts of the workflow are outside the evaluation?
Bring a workflow, not sensitive records.
An initial email can describe the process and desired evidence without sending personal or operational data. The environments use synthetic records and do not connect to live government systems.
Contact talal@maydanlabs.com.