Skip to content
Kempt

Performance

Kempt should cost nothing you can measure.

Budget

Section titled “Budget”
Metric Target How
content.js + content.css Under 100 KB (about 92 KB in 0.2.0) No framework. Icons are files masked in CSS.
Script per navigation Under 5 ms Features run only on their routes. CSS does most of the work.
Long tasks added 0 Small observer roots, and idempotent writes with no layout reads in loops
Observers alive Only scoped ones for running features waitFor disconnects; signals abort the rest
Timers and polling No intervals YouTube’s own events; a few capped startup waits (see below)

Rules

Section titled “Rules”
  • CSS comes first. Most features are stylesheet rules behind an html[kyt-<id>] attribute.
  • Animation stays on the compositor with transform and opacity where it can. Arrows rotate instead of swapping icons. Dropdowns animate grid-template-rows: 0fr to 1fr, a layout animation that runs only on click.
  • Every transition has a prefers-reduced-motion override.
  • An observer watches one container, such as the grid or the action row. waitFor falls back to document when no root is given, and disconnects once the element appears or after 10 s.
  • The page world waits for YouTube’s globals (signedIn, the guide data, icon names) with short capped loops of 100 ms sleeps. They end on success and never run in the background.
  • Repairs write only when needed: they check first, then write. A write that changes nothing still costs a style recalculation.

Checking

Section titled “Checking”

Open the browser’s Performance panel on Home, a watch page, Subscriptions and History, and record with Kempt on and off. The difference should be lost in the noise.