Record the conditions
Each test records the product, version or plan, access conditions and test date.
Use repeatable tasks
Choose tasks that reflect the intended reader’s workflow. Keep inputs and scoring criteria consistent across comparisons.
Keep the evidence
Retain prompts, representative outputs, screenshots and notes. Explain what was observed rather than inventing numerical precision.
Account for the whole task
Include setup, review and correction effort. Discuss failures alongside successful examples.
Earn the word tested
A review carries a tested label only when its tasks, conditions, date, evidence and limitations are documented.
Draft publication information · subject to completion before launch.