Discussion

Review each use separately and choose a comparator with the information available when acting. Measurement gains, user benefit, and long-term reuse need different evidence. Recording the reason and review conditions makes the decision open to scrutiny and revision.

The cases were author-coded without independent validation. Analyses used one dataset, one session, and one interface, with no untouched confirmation cohort or preregistration. Offline results do not establish real-world benefit, burden, or fairness.

Back at the screen, a person is reporting their own feelings. Making that report closer to a reference answers a measurement question. Adapting a system or keeping the profile for later each needs further evidence. PersonaGuard helps teams record these judgments and revisit them when new evidence arrives.

Limitations and Next Studies+

Limits of the present evidence

• Cases were purposively selected and author-coded after protocol development; there is no independent coding validation or source-author endorsement.
• Models were developed on one dataset with no untouched confirmatory cohort. The study and primary residual gate were not preregistered.
• One session and interface, overlapping references, and unavailable demographic and sensor-quality records limit generalization and persistence claims.
• Offline errors and calculated exposure do not establish live user benefit, harm, burden, privacy, control, or fairness.

What to test next

1. Independent analysts: code and resolve cases, document disagreements, and measure time, usability, and decision quality.
2. Prospective interaction study: compare a transparent baseline with reversible adaptation, specifying benefit, harm, burden, and exit criteria first.
3. Separate cohort and stimuli, dependence-aware uncertainty checks, and repeat-session or cross-interface measurements.