Legibility
Repeatable measurement
Render an identical large corpus
Each recorded run inserts fresh text into the same visible measurement area, then reads its geometry to force text shaping and layout. The order alternates, and every pair receives identical but newly identified text to reduce cache and first-run bias.
Ready. The benchmark can briefly load the CPU; start it when convenient.
| Mode | Computed value | Style + shaping + layout | Probe width | Rendered height |
|---|---|---|---|---|
optimizeLegibility
|
— | — | — | — |
optimizeSpeed
|
— | — | — | — |
Results will show the median of the recorded runs.
This intentionally measures synchronous style, shaping, and layout only. Page JavaScript cannot reliably time Paint or Rasterize Paint; use the isolated pages below for those events.
Inspect rendering in DevTools
Open each mode in its own tab. Start a Performance recording, then click “Re-render corpus”; scroll through the page to exercise painting of off-screen text.
Large text preview
Side-by-side output
Each column contains the same first 80 paragraphs from the measured corpus.
Speed