Methodology

This page documents how experiments in this lab are run and how the numbers on each experiment page should be interpreted. It is also a TODO list — each section below is a stub I will fill in as the experiment set grows.

Warm-up runs

TODO. Why every benchmark discards the first N runs (JIT warm-up, cold caches, layout-engine cold start). Default warm-up count and how to override per experiment.

Iteration count and percentiles

TODO. Why p50 and p95 instead of mean. How compare() picks the slowest entry as the baseline so ratios read as “X times faster than the slowest.”

User Timing API and performance.now()

TODO. Why performance.now() over Date.now(). When User Timing marks are useful for separating phases (measure vs paint vs commit) and when they are not.

CPU throttling

TODO. How DevTools 4× / 6× CPU throttling changes the picture. Whether to rely on it or run physical low-power hardware.

Browsers tested

TODO. Which browsers and versions each experiment was run in. Notes on where engines diverge meaningfully (Safari vs Chromium vs Firefox text shaping, layout containment behavior, etc.).

What this lab does NOT measure

TODO. Bundle size impact. Memory pressure under sustained load. Accessibility tree integrity. SSR/hydration paths. Be explicit about every boundary so readers can decide whether the numbers are load-bearing for their use case.

How to read a result

TODO. A short guide to the per-experiment result panel — what counts as a meaningful difference vs noise, and how confidence intervals are reported (or not).