Static fixtures only — deterministic demo data for evaluating psi-swarm without running a swarm or exposing local history.
Marketing landing → Astro overlay
https://example.com/
Moved a static marketing hero off a dynamic Worker shell onto prerendered Astro HTML.
The baseline Worker route paid SSR + hydration cost before the hero rendered. The overlay pattern prerendered the LCP block and inlined critical CSS, cutting mobile LCP p75 by ~1.1s without changing the product URL.
Metric (mobile-mid)before-overlayafter-overlayΔ
LCP p751.68s540ms-1140 (-68%) ✓
Perf score p757196+25 (+35%) ✓
CLS p750.0400.0200 (-50%) ✓
TBT p75420ms90ms-330 (-79%) ✓
Self-hosted fonts + preload
https://docs.example.com/
Removed a third-party font chain and preloaded the local WOFF2 used by the hero.
Google Fonts added two round trips and pushed render-blocking CSS. Self-hosting with `font-display: swap` and a single preload dropped TBT and stabilized LCP variance across runs.
Metric (desktop)before-fontsafter-fontsΔ
LCP p751.32s780ms-540 (-41%) ✓
Perf score p757894+16 (+21%) ✓
CLS p750.0100.0100 (0%)
TBT p75310ms120ms-190 (-61%) ✓
Edge cache rules for static shell
https://app.example.com/dashboard
Added Cache Rules for the HTML shell while keeping authenticated API routes dynamic.
Cold TTFB dominated p75 on a custom domain. Cache Rules on the static shell plus `caches.default` wrapping for GET `/` reduced repeat-view TTFB without touching authenticated mutations.
Metric (mobile-mid)before-cacheafter-cacheΔ
LCP p752.89s1.18s-1710 (-59%) ✓
Perf score p756288+26 (+42%) ✓
CLS p750.0600.0500 (-17%) ✓
TBT p75510ms260ms-250 (-49%) ✓