What actually breaks Core Web Vitals in 2026 — and how to find it
The six failure classes behind most bad LCP, TBT and CLS scores — how each one shows up, the exact fix with code, and which audit catches it.
· 9 min read
Home / blog
Blog
Engineering notes on page speed, Core Web Vitals, SEO, and the systems we build to measure them.
We built an MCP server so AI agents can run speed tests — here's how
Streamable HTTP with a stateless server, sha256-hashed API keys, and Postgres as the queue — the architecture behind our MCP server.
· 8 min read
How we run a GTmetrix-class testing stack on one small VPS
Postgres as the job queue after we deleted Redis, one Chrome at a time inside a 2 GB container, HAR files rebuilt from the DevTools log, and a deploy that is a tarball over SSH. The whole architecture, honestly.
· 12 min read
What 60 real Lighthouse runs say about what actually breaks LCP
We ran our own test engine against the 60 most-visited sites that will answer a mobile browser. The audits everyone fixes fail on fast sites too — here is what actually separates a 2-second LCP from a 5-second one.
· 11 min read
We audited 12 framework starter templates. 11 ship zero structured data.
We ran our 41-check SEO/AEO crawler over the official Next.js, Astro, SvelteKit and Nuxt starters — 85 pages. Full scores, the three checks almost everything fails, and the methodology.
· 10 min read