Methodology
Make the claim no larger than the method.
The reasoning behind bounded observation, explicit evidence, and honest status language.
Define the question before collecting support.
A test should state what would count as a pass, failure, unresolved result, or insufficient input before the result is known.
Separate observation from interpretation.
What the capture shows, what it suggests, and what it cannot establish belong in different statements.
Keep the boundary visible.
A local or synthetic result remains local or synthetic. A dated sweep remains dated. A self-administered benchmark is not external validation.
Daily AI Agents