For software engineers and technology professionals, the daily digital experience is typically insulated by high-end hardware. Working on flagship laptops, powerful mobile devices, and robust, lightning-fast Wi-Fi networks creates an optimal environment for building applications, but it also creates a dangerous illusion. This high-performance reality is often vastly disconnected from the day-to-day conditions experienced by the vast majority of end users around the world.
While developers test code on pristine fiber-optic connections, real-world users might be navigating applications on a four-year-old budget smartphone with a dwindling battery, relying on a cellular data plan that throttles speeds after a few gigabytes, or struggling with under-invested public infrastructure. Others may simply be trying to check a schedule while standing in a dead zone at the gym. Multiply these everyday hurdles by billions of people across the globe, and a profound gap emerges between software that "works on my machine" and software that actually works for everyone. This perceptual disconnect can severely distort critical business decisions, technology choices, and engineering priorities.
To help close this perception gap using objective, real-world data, Cloudflare has announced the public release of the BEACON dataset—short for Browser Experience Across Cloudflare’s Observed Network. While Cloudflare has collected telemetry data for years on behalf of enterprise customers to give them deep, bespoke insights into how real users interact with their specific sites, the company is now opening a broad, anonymized view of these performance measurements to the public.
BEACON is built from billions of real-world performance measurements captured across 10,000 of the largest websites operating on Cloudflare’s massive global network. The dataset covers every major browser engine, is updated daily within Google BigQuery, and adheres strictly to the open standards established by the community-led RUM Archive, a publicly available Real User Monitoring database. By expanding the footprint of that foundational project a hundredfold, BEACON provides researchers, developers, and industry analysts with an unprecedented vantage point on how the web truly performs across diverse browsers, hardware devices, and nations.
What BEACON Reveals About Global Web Performance
For years, Core Web Vitals have served as the de facto industry standard for measuring perceived web performance. BEACON reports all three core metrics, but rather than stopping at a single average or standard benchmark like the 75th percentile, the dataset publishes full histograms. This granular approach allows researchers to examine any percentile they choose, diving deep into the long tail to identify precisely where the technology industry continues to struggle to deliver fast, reliable experiences for every user.
Analyzing performance across operating systems and browser engines yields immediate insights into systemic discrepancies. For instance, WebKit—currently the sole browser engine permitted on iOS—generally demonstrates strong performance metrics overall. However, that performance advantage is far from universal. In 46 different countries where WebKit accounts for more than ten percent of incoming web traffic, its Largest Contentful Paint (LCP), Interaction to Next Paint (INP), or both metrics trail behind Blink-based browsers such as Chrome, Edge, and Opera by at least ten percent. In Cambodia, where WebKit represents roughly 17.5 percent of page views, its LCP performance is 50 percent worse than that of Blink-based alternatives.
BEACON also integrates domain industry classifications sourced from Cloudflare’s Intel API, revealing stark performance contrasts across different sectors of the internet. Government and politics, health, and child-friendly categories consistently emerge as high-performing sectors, whereas advertising, religious, and weather-related domains frequently rank among the slowest. Furthermore, the inclusion of extended percentiles exposes hidden vulnerabilities that traditional single-metric reporting often obscures, particularly concerning visual stability as measured by Cumulative Layout Shift in the long tail.
Pinpointing What Makes Pages Feel Slow
To help developers diagnose performance bottlenecks, BEACON extends the RUM Archive standard by breaking down LCP and INP into specific sub-parts that isolate each individual stage of loading and response. These diagnostic metrics will also be integrated directly into Cloudflare’s native Real User Monitoring dashboards in the coming weeks. By categorizing performance into standardized thresholds for good, needs improvement, and poor, engineering teams can pinpoint exact areas for optimization.
When examining LCP sub-parts—which measure Document Time to First Byte, load delay, load duration, and render delay—the results challenge a common assumption among developers. Conventional wisdom often dictates that downloading heavy resources like images, fonts, or videos is the primary culprit behind sluggish loading times. However, the data reveals that downloading the resource itself typically contributes the least to perceived loading delay. For most page views that breach performance thresholds, the more significant opportunities lie in accelerating resource discovery and unblocking the rendering process. Cloudflare customers can tackle resource-discovery bottlenecks using features such as Smart Hints.
Similarly, the dataset breaks down Interaction to Next Paint into input delay, processing time, and presentation delay. For the slowest interactions, JavaScript execution time accounts for the longest periods, accompanied by a noticeable rise in presentation time, which is typically driven by complex cascading style sheet layout recalculations. Tools such as Zaraz can assist developers in mitigating the performance drag caused by heavy third-party JavaScript integrations.
How Application Architecture Changes the Picture
The impact of modern application architecture on web performance is another major focus of the BEACON dataset. Cloudflare recently added native support for Google Chrome’s Soft Navigations API, which enables accurate LCP reporting for client-side navigation models commonly utilized in single-page applications built with popular frameworks like React, Vue, Angular, and Svelte.
Comparative analysis shows that soft navigations render two to three times faster than traditional hard navigations across every measured percentile. However, this architectural efficiency does not eliminate the performance cost of the initial landing page, which tends to be considerably heavier. Development teams adopting single-page application frameworks must carefully weigh these architectural tradeoffs to ensure that faster subsequent navigations adequately offset a slower initial user experience. If visitors rarely progress beyond the landing page, the heavy initial load may never yield a net performance benefit.
Combining Data Sources for Deeper Insights
Because BEACON is structured as an open dataset, its analytical value extends far beyond its native fields. Researchers can seamlessly join it with external macroeconomic and infrastructural datasets to uncover complex correlations. For example, cross-referencing BEACON with World Bank data on gross domestic product per capita reveals direct correlations between national economic conditions and regional web performance.
Cloudflare Radar, the company’s public data insights and visualization platform, will incorporate this methodology into a dedicated web performance section. This new section will feature direct correlations between Cloudflare’s Internet Quality Index data and the BEACON dataset. The IQI aggregates performance benchmarking telemetry gathered during the company’s bi-annual network performance updates.
Pairing BEACON with the IQI is particularly powerful because it separates the user experience into its two most influential components: the performance of the specific website being visited, and the performance of the local internet service provider network delivering the connection. Both factors heavily influence page load speeds, and together they determine whether a web browsing session feels seamless or frustrating. Early analyses show that these two variables generally move in tandem; high network bandwidth typically coincides with faster perceived load times, though regional anomalies—such as smaller average transfer sizes observed in Africa—suggest that local businesses are actively adapting their web content to optimize for constrained network environments.
Data Processing, Privacy, and Accessibility
Releasing a real-world performance dataset of this magnitude requires rigorous safeguards to protect user and corporate privacy. Cloudflare’s underlying Real User Monitoring product was engineered from the ground up to be privacy-first. For the BEACON dataset, the company strips out all customer-specific website identifiers, including domain names and granular uniform resource locator paths. This ensures that the global research community gains actionable insights without compromising the trust of the individuals and organizations contributing to the network.
To prevent skewing the data, Cloudflare addressed the challenge of managing millions of distinct domains by establishing a normalized scope focused on the 10,000 largest websites across its network. This sample size strikes a careful balance, providing global diversity across technological architectures, regional footprints, and traffic volumes while collectively representing billions of daily records. Furthermore, records sharing dimensions such as country, operating system, browser, and connection protocol are aggregated, and any groupings with fewer than five data points are discarded to completely eliminate the risk of individual or site re-identification.
The BEACON dataset is publicly available today on Google BigQuery, accompanied by documentation and sample queries that developers and researchers can adapt for their own analyses. By making this unprecedented level of geographic and browser diversity accessible to the public, Cloudflare hopes to empower browser vendors, standards organizations, and developers to identify where the modern web falls short and help build a faster, more accessible internet for everyone. The launch of the dataset was made possible through the contributions of participants in Cloudflare’s intern program, highlighting the collaborative effort behind this milestone project.
Leave a Reply