How to be a Performance Detective?
summary
Walks through real Etsy casefiles that connect Core Web Vitals regressions to code via webpagetest waterfalls, synthetic Lighthouse runs and in-house RUM piped into A/B experiments. Shows a 4.5% p75 LCP regression traced to an extra CSS file pushing the LCP image later, a CLS spike caused by a mismatched skeleton placeholder height, and an elusive FID blip with inconclusive attribution. Uses CDFs for RUM and Lighthouse avoid large layout shifts to diagnose what changed and where.
AI-written summary · about
chapters (13)
- 0:00Performance detective intro
- 1:26Detective Pikachu theme
- 3:36Hard truths on web perf
- 5:40Metric types and matchups
- 8:11Field vs lab metrics
- 10:42RUM data visualization
- 12:52Synthetic monitoring tests
- 14:17Primary vs diagnostic metrics
- 19:29Case study LCP degradation
- 26:34Debugging CLS shifts
- 30:27FID regression challenges
- 36:53Lessons for detectives
- 38:41Q&A
related talks
49mDemystifying web performance tooling and metrics
43mDemystifying and leveraging Web Vitals
27mHow to Measure Performance Effectively?
31mPanel Discussion: Get to Know: Web Performance and Core Web Vitals
15mWhat's new in Web Vitals
16mFrom Beginner to Expert in Web Performance
01h 14mCore Web Vitals: Theory and practice
50mHow to Think like a Performance Engineer