A credible evaluation begins with a question that can be answered. State the intended scope, input conditions, evaluation data, exclusions and decision threshold.
Describe the conditions
Record the sensor, capture conditions, time window, compute environment and changes made during the test. If the sample is small or unrepresentative, say so.
Report errors and uncertainty
Detection counts alone are rarely enough. Include false positives, missed cases, ambiguous examples and data limitations where the evidence allows.
Keep claims proportional
A controlled experiment supports conclusions about those conditions. It does not establish performance in another site, season, population or environment. Preserve limitations alongside the result.