Evaluation identity
6 guided items
Pilot, build and managed operations
LLM Evaluation Template
Representative cases, acceptance criteria and regression evidence.
How to use this resource
A benchmark should represent the actual workflow, users, data, exceptions and harms - not only general model performance.
Important: General technical template; evaluation results do not establish absolute accuracy, security or compliance.
Case taxonomy
6 guided items
Measures
5 guided items
Failure review
5 guided items