Performance Engineering
Field notes on Core Web Vitals, real-user data, and frontend performance at the session level.
Why Lab Scores Lie While Users Bounce
A Lighthouse score of 97 feels like a win — until you check your real-user p75 LCP on mobile and see 4.2 seconds.
p75, Not p50: The Percentile That Actually Predicts Bounce
The median hides your problem users. If your p75 LCP is 4 seconds, one in four page loads is slow.
Third-Party Scripts: The Hidden Render Blocker Audit Teams Miss
Your first-party code is fast. Your third-party tag manager, consent platform, and analytics suite are collectively adding 1.4 seconds to your LCP.
RUM vs. Synthetic Monitoring: When Each One Actually Wins
They are not competitors — but most teams are using the wrong one for the wrong job.
Mobile Web Performance: Cold Starts, Cached Views, and What Your p75 Hides
A cached second visit and a cold-start first visit behave like entirely different pages.
CLS Deep Dive: Why Layout Shifts Kill Conversions, Not Just Scores
A CLS of 0.12 is technically a 'needs improvement' score. But the actual experience is a button that jumps as they tap it.
Segmenting Real Users by Network, Device, and Geography
Your p75 LCP on desktop Chrome in Seattle tells you almost nothing about your users in Mumbai on a mid-range Android phone over 4G.
Alerting on Performance Regressions Without Alert Fatigue
Deploy-correlated p75 alerting catches regressions in minutes. Here is how to set thresholds that fire when it matters.