Measurement practice in Core Web Vitals research: A mapping of 357 works

Abstract

Core Web Vitals have become the reference metrics for client-side web performance, and a literature reporting them has grown from five works in 2020 to 152 in the first ten months of 2026. The measurement practices inside that literature have not been described, so what its accumulated evidence can support is not known. This systematic mapping study examines what the literature reporting Core Web Vitals metrics actually measures, asking which data sources the measurements come from, which metrics are reported, whether connection quality enters the analysis, and which populations and author communities the work represents. Seven precise phrase queries were run against the OpenAlex registry on 25 October 2026, unioned and deduplicated, yielding 460 records. Six screening gates reduced these to an analytic corpus of 357 works published from May 2020 onward. Each work was coded on six dimensions against rules fixed before coding began, with geography coded twice over and the two measures kept apart. Lab tools that simulate a single visit are named by 136 works and real-user or archive sources by 16. Of the 266 works reporting measurements, 123 name no data source at all. Connection quality enters 56 works of 357, and the pattern inverts by source, since synthetic studies consider it in 60.7 per cent of cases while the six works using the Chrome User Experience Report consider it in one. Lighthouse cannot measure Interaction to Next Paint, and among the 136 lab-tool works that metric appears nine times against 61 for the Total Blocking Time proxy. Only 15 works of 357 report all three currently stable Core Web Vitals together. Of the 207 works carrying affiliation data, 84 have an Indonesian institution and eight an African one. The field setting measurement norms for the web measures that web with simulated single visits, frequently without saying where its numbers came from, and largely without the connection variable its own instruments record. Reviewers and editors can close the traceability gap at low cost by requiring a named data source.