Proceedings of the International scientific and practical conference ―Synergy of Modern Science and Education‖ (February 2-4, 2026) / Publisher website: www.naukainfo.com. – New York, USA, 2026. - 324 p.

76  The reliability of eye metrics depends on lighting, posture, and camera quality; therefore, in practice it is worth having graceful degradation to behavioral metrics when video signal quality is insufficient [9];  Blink rate has ambiguous correlations with load in different contexts; it is more correct to interpret it as an additional cue rather than a primary indicator [17];  Privacy: the camera is a sensitive data channel. For educational products, transparency (explicit user consent), minimization of data collection (local processing without sending video to a server), and the ability to learn without a camera (via interaction metrics) are critical. The user must have full control over camera usage. Legislation, including the GDPR and national standards for biometric data protection, imposes strict requirements on processing video content, which calls for a privacy-aware system architecture [18];  Correct formulation of purpose: the system supports the learning process and does not diagnose an individual‘s psychological state. It is important to avoid stigmatization (―the system thinks you cannot cope‖) and to present adaptation as help rather than an evaluation;  Transparency of logic: the user should understand why the interface has changed and should have the option to manually return to Normal/Power mode if the automatic adaptation is incorrect. Conclusions. The paper proposes the concept of moment-aware adaptive e- learning, which combines two lines of personalization: (1) the learner‘s long-term profile and (2) the current in-the-moment state. The theoretical foundation is Cognitive Load Theory, which emphasizes the role of instructional design in reducing extraneous cognitive load. As a practical adaptation mechanism, progressive disclosure and a mode-based interface design system (Focus/Normal/Power) are proposed, which change the presentation of material and UI complexity depending on overload signals.