A readable health interface is not a clinically validated system
Published October 7, 2026 ยท Independent editorial guide
A system can perform well on one evaluation and still leave its most important claim untested. Physiological-computing researchers know this problem when comparing signal-processing results across participants and sessions. The site's historical benchmarking page supplies the research context. This new note extends the question to information interfaces, without claiming a new PhyPA study or a medical application of its historical datasets.
State the task being evaluated
Finding a document, understanding its wording and making a clinical decision are different tasks. A usability exercise might ask whether a reader can locate a source date. That exercise cannot establish whether a device diagnoses a condition or whether an educational intervention improves health. Write the task, participant group and outcome before choosing a measurement.
Use an information site as an interface example
The historical New Mexico diabetes education archive groups professional education, patient education and community programs. Its 2009 material can serve as a small example for examining navigation labels and historical-context cues. It is not a physiological dataset, a diagnostic tool or a currently validated intervention.
A bounded task could ask readers to identify which section appears intended for professionals and to find the archive's date notice. Record completion, uncertainty and the wording that caused confusion. Do not ask participants to supply a diagnosis to perform this task, and do not infer their condition from which link they select.
Keep comparison conditions visible
Record page version, device, language, font scaling, connection failures and the exact task instruction. Changing several of these between sessions can alter results independently of the navigation change under study. Distinguish a participant who cannot reach a page from one who reaches it but cannot identify the relevant information. Missing observations should not become successful trials or zeros by default.
Report what the evidence does not support
For present-day general health reading, use a separate source such as NIDDK's diabetes information. A comparison with that source would still need a clearly defined task and could not turn a usability result into a treatment recommendation. A defensible report states the tested interface question, the observed behavior and the remaining limits. Clinical validation requires evidence appropriate to the clinical claim, not a borrowed benchmark score.