What should I ask for in an escalation sample or pilot?
A checklist of concrete items to request in a sample, so you can judge consistency, difficulty and fit with your agent before any wider engagement.
Frame the request around what you need to inspect and run, then ask for the following.
- A small set of tasks per trigger type, each labelled with its trigger and difficulty. - The matching records: cases, parties, facilities, reviewers, policies and an authority matrix with dates. - An expected-outcome block per task, naming the correct target role and required handover facts. - At least a few cases that should not be escalated, and a few that are ambiguous. - A manifest with schema versions, key relationships and generation settings. - A validation script or description for referential integrity and date order. - Your own role names and field names mapped in, where practical. - A written note on how the material was generated and what to review for grounding and disclosure risk.
Decide beforehand which result would make you proceed, for example your agent handling a stated trigger reliably, and which would make you stop. Share that with MirrorFirm so the sample is built to test it. Quantities, format and terms for the sample are agreed in discussion.
A well-specified sample request lets you reach a go or no-go decision from evidence you have inspected yourself.
- Audience:
- Agent developers
- Content type:
- Checklist
- Buyer question:
- Scope and getting started
Content release: r-a220eb6e099b243b
Related questions
- What licensing questions should I raise for generated escalation material?
- What might a first escalation pilot show an agent team?
- What should my team prepare to start an escalation pilot?
- How should I scope a first escalation pilot with MirrorFirm?
- What licensing, limits and boundaries should a buyer ask about before a pilot?
- What should a pilot package manifest and loader contract specify?