← Research

Trailhead: 6 Builds, 1 Site

Same 3-page site. Six stacks. What each one actually costs — and what a crawler actually sees.

What was tested

The same 3-page hiking-trail site, built six separate times: two vanilla approaches (vanilla-static, a fully static multi-page build, and vanilla-spa, a client-side-routed single-page build), a custom Rust backend (site-server, real per-request server rendering), and the three mainstream frameworks — React, Vue, and Angular. Every stack implements the same four features — routing/transitions, a component, shared state, and a form + list — and every measurement below comes from each stack's actual production build (npm run build / cargo build --release, or used as-is for the two vanilla builds) served over real http://, not file:// and not a dev server.

What was found — deployable footprint

The three vanilla-ish stacks (vanilla-static, vanilla-spa, site-server) all ship 27–28KB of files a browser ever has to fetch for the same 3 pages. Vue ships 110.7KB, Angular 242.5KB, and React 293.0KB — React alone ships roughly 10x the browser-facing bytes of any vanilla stack for identical pages and features. (site-server's 1.55MB Rust binary lives on the server and is never downloaded by a visitor — it doesn't count against the browser-facing number.)

Dev-time tooling weight tells a similar story: the vanilla stacks need zero installed packages. Angular's node_modules alone comes to 210.1MB across 577 lockfile entries — bigger than roughly 7,000 copies of the entire vanilla-static site.

The standout finding — SEO/GEO crawlability

The sharper result isn't about bytes at all. Search crawlers and most LLM crawlers/retrievers (GPTBot, ClaudeBot, PerplexityBot, CCBot) fetch raw HTML — they don't execute JavaScript. Checked directly with curl against each stack's production build: React, Vue, Angular, and vanilla-spa all serve one generic <title> and zero of the seven trails' content in their raw HTML, for every route.

That's 4 of the 6 builds functionally invisible to a non-JS crawler. Only vanilla-static (real per-route files, hand-authored) and site-server (real per-request server rendering, with robots.txt/sitemap.xml/llms.txt generated at startup so they can't drift) actually deliver real per-page titles and content to anything that isn't running a browser.

Browser-facing footprint and non-JS crawler visibility, by stack

Stack	Browser-facing size	Visible to a non-JS crawler
vanilla-static	27–28KB	Yes
vanilla-spa	27–28KB	No
site-server (Rust)	27–28KB	Yes
Vue	110.7KB	No
Angular	242.5KB	No
React	293.0KB	No

The trap

The four "invisible" stacks aren't skipping SEO on purpose — each one got a real code change that updates document.title and the meta description client-side (Angular's Title/Meta services, Vue's router meta + afterEach, the others directly in a hook). It works, in the sense that a real visitor's browser tab updates correctly. But a plain HTTP crawler never runs that code, so it receives the exact same generic title and empty content it did before the "fix" — the change only ever reaches people, never crawlers.

Honest caveats

Not every number favors the vanilla stacks. On localhost — near-zero network latency — the framework builds had faster DOMContentLoaded despite shipping far more bytes, because they bundle everything into a single JS file (3 requests total) while the vanilla stacks split logic across 6–7 small files (8–10 requests). On a latency-bound network more requests cost more; on a bandwidth-bound network more bytes cost more — neither number alone settles which approach is "faster," and it depends on the network the site actually runs on.

Lines-of-code told an inverse story too: Vue needed the least hand-written glue for the four shared features, and the vanilla stacks needed the most, since they implement their own router, store, and component system instead of importing one.

The full 9-page comparison deck below covers all of this visually, including the raw data tables — page 9 is explicitly the author's personal case for the Rust server, marked as such, not the deck's neutral verdict.