Files
AndreaSlopFotografie/aidlc-docs/features/soorten-fotoshoots-page/construction/build-and-test/performance-test-instructions.md
T
SluijsensandClaude Opus 5.5 a4149e86af
Continuous Integration / check (pull_request) Successful in 10m46s
Continuous Integration / deploy-test (pull_request) Skipped
Continuous Integration / lighthouse (pull_request) Successful in 12m15s
Continuous Integration / deploy-production (pull_request) Skipped
Records Build and Test of the Soorten fotoshoots page and completes the feature
Full pnpm check at d12665c passed (325 tests, Lighthouse within budget)
and 19 browser checks against the production build passed. Adds the
build and test instructions and summary, marks the feature complete
and logs the session in the audit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-25 16:30:33 +01:00

1.9 KiB
Raw Blame History

Performance Test Instructions — soorten-fotoshoots-page

A static site, so there is no load testing. Performance means page-load performance, measured with Lighthouse CI on the production build.

cd web
pnpm build
pnpm lighthouse       # mobile (lighthouserc.cjs) then desktop (lighthouserc.desktop.cjs)

Before running: send "claim 4173" to the parallel feature sessions and keep the machine quiet; afterwards "4173 free".

Budgets

  • Performance, mobile: homepage ≥ 0.8 (NFR-P-01); photo pages /over-mij and /soorten-fotoshoots ≥ 0.85 (new in this feature); every other page ≥ 0.9
  • Performance, desktop: every page ≥ 0.9 (lighthouserc.desktop.cjs now raises every entry, not only the homepage's)
  • Accessibility and best practices: 1.0 on every page
  • SEO: 1.0 on the indexed pages; the basic audits on noindex pages

Measuring reliably

Check environment.benchmarkIndex in the JSON report: around 3900–4300 on this PC. One homepage run at 0.69 had a benchmarkIndex of 1909 (the machine at half speed during that measurement); such a run says nothing about the code.

Why the photo pages get 0.85 on mobile

Lighthouse simulates a slow mobile connection (lantern). On a page whose largest paint is a photo, that photo can only be requested after the main bundle and the page's own chunk have run, so /over-mij and /soorten-fotoshoots both land on LCP 3.22 s and a score of about 0.897 (FCP 0.72, SI 0.98, LCP 0.72, TBT 0.99, CLS 1.0). That rounds to 0.89 or 0.90 from run to run. Tried during this feature and measured: fewer bytes on the LCP photo (848×1060 instead of an upscaled 960×1200) and no fade-in on the priority photo; neither moved the LCP. Moving the page's JSON-LD out of head() kept the main bundle 1 kB (gzip) smaller. The structural fix is prerendering or a smaller main bundle (NFR-P-01); then the photo pages go back to 0.9.