From research question to experiment: a one-page design template

Before collecting data, answer five questions on one page: what phenomenon should be explained, which variable changes, what is the comparison, how will the result be judged, and where might it fail? If these are unclear, more data and larger models usually add noise.

Five blocks on one page

Write the question and a falsifiable expectation. Then write variables, samples, and exclusion rules; baselines, controls, and treatment groups; primary and secondary metrics with the statistical method; and finally risks, resource limits, and stopping conditions.

Return to the page after the experiment and mark which settings changed and which results did not support the original expectation. This separates the original design from a post-hoc explanation.

Start with a minimum experiment

Choose the smallest experiment that can test data availability, metric validity, and pipeline execution. It may not prove the hypothesis, but it can expose definition and engineering problems early.

A design template does not replace domain knowledge, statistical judgment, or ethics review. Follow the formal process for studies involving people, private data, or high-risk applications.