Performance (measured)

Real numbers, not a hunch — measured on this deployment. bext's interactivity floor (the JS a client pays to make an island interactive) is in the single-digit kilobytes, and zero before interaction in lazy mode.

Interactivity JS, compressed (gzip)

What a client downloads to make a counter interactive. The resume runtime (~3KB) is cached once across every island and page; each extra island adds only ~0.6KB — or nothing until interaction.

React + ReactDOM45 KB

SolidStart (solid-js)7 KB

bext — island + shared rt3.5 KB

bext — each extra island0

bext — lazy, before interaction0

Tip

This is the framework floor, not a site's total. A heavy site (including this demo template, with its syntax highlighting and embedded editors) ships its own chrome on top. The point isn't that every bext page is tiny — it's that the framework doesn't force a heavy baseline on you: the floor is ~13× lighter than React's ~45KB (and each extra island ~75×), and zero before interaction.

Head-to-head: the same app in five frameworks (measured)

We built the same app — a server-rendered 24-product grid + counter, theme toggle and search filter, all verified in a real browser — in all five frameworks, then forced each to render fresh on every request (no cache, no prerender; proven by a timestamp that changes on every request). Production builds and servers on activ-2; RPS/TTFB via oha -c 50.

Framework	Client JS (gz)	TTFB	RPS (uncached)	p99	Build
bext (PRISM)	5.6 KB	6.0 ms	4,742	71 ms	~4 KB*
Astro	0.33 KB	12.3 ms	3,807	20 ms	432 KB
Remix	101 KB	12.2 ms	1,011	75 ms	335 KB
Qwik	35 KB	42.6 ms	938	73 ms	249 KB
Next.js	145 KB	21.1 ms	575	141 ms	5.2 MB

Tip

Even uncached, bext renders fastest (Next's force-dynamic dropped it from 2,945 to 575 RPS — it had been serving prerendered static, not rendering). With its built-in ISR cache — one config line, not a CDN — the same route serves 282,776 RPS at 0.13 ms TTFB. The uncached 4,742 is bext's worst case, and it still wins. * bext has no heavyweight build: it compiles on demand.

Time to First Byte

Rust render + ISR cache. The full resumability page leaves in ~18ms (first-byte and total are the same over the local network), across three warm requests.

~18 ms

Methodology (reproducible)

reproduce.shShell

# interactivity floor — gzipped bytes a client actually pays
curl -s https://demo.bext.dev/islands/ResumableCounter.js | gzip -c | wc -c   # 589
curl -s https://demo.bext.dev/islands/__resume-runtime.js  | gzip -c | wc -c   # 2970 (cached once)

# TTFB (Rust + ISR cache) — three warm samples
for i in 1 2 3; do curl -s -o /dev/null \
  -w '%{time_starttransfer}s\n' https://demo.bext.dev/examples/resumability; done
#  → 0.018s  0.020s  0.019s

# full harness (TTFB + JS bytes + request count per page)
node harnesses/browser/bench.mjs