[harlan-zw]

❓ Type of change

• 📖 Documentation
• 🐞 Bug fix
• 👌 Enhancement
• ✨ New feature
• 🧹 Chore
• ⚠️ Breaking change

📚 Description

The homepage's first frame waited for the whole page to hydrate. In a Lighthouse trace of the mobile run, the first paint came right after the hydration task, about 650 ms after the HTML had parsed. Lighthouse then charged every preloaded script to the hero heading, the page's LCP.

The demos, the trending board and the agents band now use Nuxt's hydrate-on-visible. They still render on the server with every link and text, and their code loads and hydrates once they scroll into view. The hero, the lifecycle band and the sections that fetch their own data hydrate as before.

Five alternating Lighthouse 13.5 mobile runs of each build, cold cache, local pnpm build served by wrangler dev, medians:

Metric	Before	After
Observed first paint	2,353 ms	746 ms
LCP, simulated	7.2 s	3.9 s
FCP, simulated	5.3 s	3.5 s
TBT	440 ms	241 ms
Score	51	76
modulepreload links in the HTML	43	38

The machine ran other builds at the time, with a load average of 44 to 73, so single runs vary. The builds alternated run by run to share that load.

🤖 AI disclosure: Harlan Agent Kit modified this description. My AI open-source policy.

[harlan-zw]

Agent check (Claude Code, Harlan Agent Kit), since CI cannot cover it:

• Compared each element's box and computed style between main and this branch on the six launch pages, at 412 and 1350 pixels, with scripts off and on: 0 rows differ.
• On the local build, a demo pick on the homepage still switches the stage after the section scrolls into view, at both widths, with no console errors or warnings.