Make performance part of publishing
Optimizing each image or font by hand creates work to repeat with every update. We include asset optimization and quality measurements in the publishing workflow so the same checks continue as articles are added.
Delivery & measurement
Separate initial delivery from reuse
Combine build-time optimization, cache policies and measurements for each change.
Prepare before release: build smaller files
Astro generates HTML in advance, while CSS and JavaScript are minified. Shared CSS is included in the HTML to avoid an extra request for styling. Reading an article does not query a content database.
The build generates WebP images at several sizes. The browser selects a version suited to the display width and pixel density. Explicit width and height reserve space before images load.
The font used on Japanese corporate pages is subset to the characters in public content and UI, then served from this website. Article listings and article pages use native device fonts to avoid an extra font download when reading.
Reuse: keep content current without downloading everything again
HTML and fixed-URL CSS and JavaScript are revalidated before being fetched or reused, preventing old content and styling from remaining indefinitely.
Fonts and generated images use filenames that change when their content changes. Because the same URL is not overwritten, these files can use a one-year cache with immutable. Changed files receive new URLs, while unchanged files remain reusable.
Measure every change: make checks repeatable
For each pull request, CI serves the deployment artifact on a local Worker and runs Lighthouse three times per Japanese and English page under both mobile and desktop conditions. Median results and HTML reports make the impact of a change available for review before release.
These are lab measurements intended for repeatable comparisons. They differ from PageSpeed Insights measurements of production URLs and from real-user data. Production CDN behavior, devices and networks still need separate assessment.
Connect metrics to the reading experience
A score alone does not describe the experience. These metrics help us examine what happens while someone reads or interacts with a page.
- LCPLargest Contentful Paint
- How long it takes for the main image or text to appear: the wait before the page becomes useful to read.
- INPInteraction to Next Paint
- The response time from a click or tap to the next visual update: whether interactions keep the reader waiting.
- CLSCumulative Layout Shift
- How much content unexpectedly moves during loading: whether text or a button shifts while someone reads or tries to use it.
A standard Lighthouse navigation audit does not measure INP throughout a visit. We need interaction checks and real-user data alongside it.
Diagram connections animate only in view, with a pause control and support for reduced motion. Analytics loads only after consent on production, and contact and completion pages have no GTM. Performance work should support reading and interaction.
