pagespeed: fix in lab, judge in field, retest right

web · 2026-10-13 · 8 min

chasing 100 on pagespeed while customers bounce on a 70 is the industry's favorite category error. lab (simulated moto-G, throttled) vs field (real chrome users, 75th percentile) measure different things — google ranks and rewards the field. the distinction tool-led guides (sematext tables, cloudflare explainers) bury under checklists, plus the retest discipline.

lab vs field (the only distinction that matters)

lab: reproducible, simulated, debuggable — use for diagnosing (which asset blocks? what shifts?). field (CrUX): real users, real networks, real phones — use for judging (are visitors actually fast?). pagespeed shows both; owners read the lab number and ignore the field verdict. rule: fix in lab, judge in field. field green + lab orange = ship it. lab green + field red = your simulation lies about your audience.

reading YOUR report (decision tree)

no field data (low traffic — CrUX needs volume): use lab + web-vitals JS snippet; treat scores as directional. LCP red: hero image late (preload, compress, responsive sizes — 73% of mobile LCPs are images per web almanac). INP red: JS hogs the thread (defer, split tasks, drop widgets). CLS red: unsized media/ads/fonts (dimensions, placeholders, font-display). TTFB red: server/hosting (the floor — caching, CDN, better host). symptom → evidence → ONE fix → retest: the triage rivals organize as flat lists.

the retest discipline (what nobody teaches)

one change, one retest (parallel changes teach nothing). field lags weeks (CrUX aggregates 28 days — today's fix graduates next month; patience is methodology). lab variance ±10 (three runs, median — single runs lie). money pages only (homepage, top service, checkout — perfect blog-archive scores pay nothing). monitoring over scores (alerts on field regressions beat quarterly 100-hunts). perfect is a hobby; green-field is business: speed sequence. diagnosis: why slow.

FAQ (from PAA)

what is a good pagespeed score?

field: all-core-vitals green at 75th percentile. lab 90+ nice, 100 unnecessary. score without field context is arcade, not engineering.

which tool: pagespeed, GTmetrix, pingdom?

pagespeed for google's verdict (field + lab + opportunities). GTmetrix/pingdom for waterfalls and history. one verdict tool, one diagnostic tool — more tools, more confusion.

if-this-then-that diagnostic

lab score is not the core-web-vitals assessment. diagnose in lab, prioritize by field, change one thing, retest, watch trend. speed work: speed up. images first: optimize images.