Freeze the setup.
Record the AI tool or model version, the task, workflow, data and relevant settings. Make the conditions clear and record material changes between runs.
HOW WE WANT TO WORK
The evidence.
And its limits.
A result is more useful when you can see the conditions behind it. This is our proposed publishing standard.
Define the question.
Define the research question, comparison, evidence and success criteria before collecting results. Assess practical usefulness alongside quality, cost and reliability.
Repeat the experiment.
Use fixed cases, preserve failures and report the number of runs. Keep raw outputs available when privacy and licensing permit.
Report what happened.
Show observations and uncertainty alongside the conclusion. Distinguish an early idea, a working prototype and a finding supported by repeated evidence.
Make the limits visible.
Explain what the experiment cannot establish, what would invalidate the result and how a reader could reproduce or challenge it.
A CLEAR DISTINCTION
Plans are not findings.
Observed findings describe their conditions and limitations. Sample study plans remain labelled as plans and do not establish a result.
Explore findings and study plans