Experimental Design Guard
Design errors are the hardest category of error to catch after the fact — because the analysis looks correct given the data structure, and the data structure is wrong. The Experimental Design Guard uses declared metadata (subject IDs, batch IDs, timepoints) to surface common design-level risks.
What is Experimental Design Guard?
Design-level errors — pseudoreplication, batch confounding, ignored within-subject correlation — cannot be corrected at the analysis step. They require either re-analysis with the correct model or honest acknowledgment of the limitation. The Experimental Design Guard surfaces these risks early, when there is still an opportunity to review the design assumptions. See each page for current product behavior and known limitations.
Design errors are not violations of statistical tests — they are violations of the independence assumptions that all standard tests require. A t-test run on pseudoreplicated data does not produce an error message. It produces a p-value that is more significant than the data warrant, with no warning.