Evidence with
clear boundaries.

Measured is different from intended

A design intention must not be labelled as an established capability. An experiment must not be presented as a production guarantee.

Claims need a traceable record

Link each published result to its product revision, test definition, inputs, outputs and review decision.

Limitations belong beside the result

Describe what was not tested and where the finding is not expected to generalise. Avoid collapsing unrelated capabilities into a single confidence score.

Correction is part of the process

Supersede invalid results transparently. Preserve the reason for a change rather than silently replacing the history.