How We Test — Benchmark Methodology

Reproducibility is the whole point

Most "hosting reviews" report numbers you can't reproduce. We do the opposite: every metric is a timestamped row in our database (metric, value, testLocation, deviceProfile, capturedAt) — a real time-series you can audit. We publish the formula and the cadence, and we link this page from every benchmark chart.

Canonical test sites

Wherever possible we benchmark an identical small WordPress install (same theme, same fixed content) hosted on each provider — apples-to-apples. Where a provider has no canonical test site yet, we measure their reference site and label it clearly as "reference site / no test site" rather than pretending it's comparable.

What we measure, and with what

• PageSpeed Insights (Google Lighthouse), mobile + desktop — TTFB, FCP, LCP, TBT, CLS, Speed Index, and the overall PSI score.
• PSI CrUX field data — real-user LCP & CLS where available.
• Cloudflare Browser Rendering (our own edge) — navigation-timing TTFB and fully-loaded.
• Uptime — our own lightweight checks every 5 minutes → rolling 30/90/365-day uptime %.

Scheduled, multi-source, edge-run

Tests run on a schedule (cron) inside Cloudflare — Browser Rendering + the PageSpeed Insights API — and write to our time-series store. Anomalies are flagged deterministically (a point above 2× the trailing median). Screenshots are real captures only, never AI-generated.

See the freshness for yourself

The data-freshness page shows the last benchmark/scrape per provider from live timestamps, and the scoring page shows exactly how the numbers become a score.