Longitudinal user experience research depends on capturing how perception evolves over time. Two practical problems limit current tools: participants reconstruct missed reports after the fact (“back-filling”), contaminating in-situ datasets with retrospective fabrications; and the qualitative volume produced by long studies exceeds what manual analysis can absorb. We present UXTracker , a web-based instrumentation platform engineered for longitudinal UX studies. The system makes three architectural commitments. First, all temporal and protocol decisions are taken on the server, so the client cannot bypass cooldowns or alter timestamps. Second, the data model separates the semantic definition of evaluation aspects from the storage of raw measurements, allowing researchers to add or modify evaluation dimensions during a study without database migrations. Third, qualitative interpretation is generated by a Large Language Model that operates strictly on a structured analytical context derived from server-side clustering, never on the raw dataset. We deployed UXTracker in a five-wave field study with 51 participants, distributed across in-situ and retrospective protocols, and collected 765 evaluation points on a third-party learning platform. The deployment demonstrates that the architecture concurrently enforces two temporal regimes within a single study, preserves the granularity differences expected between protocols, and produces grounded narrative interpretations.
Assunção et al. (Mon,) studied this question.