The CSS subset

There is no cascade and no layout engine. Every element has a CSSStyleDeclaration that stores and reads back what you set — cssText, setProperty, getPropertyValue, removeProperty, getPropertyPriority, length, item, and camelCase properties for the common names, kept in step with the style attribute — but only a backed element parses the subset and does anything with it.

Backed means <canvas>, <video> and <iframe>. Those three become compositor layers; nothing else has a box.

The subset

Property	Notes
`position`	what makes a rect a rect
`left`, `top`, `right`, `bottom`	px and %
`width`, `height`	px and %
`z-index`	sorts canvas layers and instance layers in one list
`opacity`	applied by the compositor’s alpha blend
`display: none`	hides the layer without releasing it
`transform`	`translate` and `scale` only

An accepted value reads back normalised, the way a browser does it — '0' reads as '0px'. A value outside the subset is not applied, and warns once.

hud.style.cssText = 'position: absolute; left: 10%; top: 10%; width: 50%; height: 50%; z-index: 2'
hud.style.left                 // '10%'
hud.getBoundingClientRect()    // the placed rect, in drawable pixels

On a canvas, the subset decides what the canvas is

A canvas the subset has placed, resized or hidden before it asks for a context becomes a layer at that rect. One it has not takes the page’s frame, if the frame is free. See Canvases and WebGL.

Values that merely match the drawable place nothing and say nothing — Lightning’s style.width = '1920px' on a 1920-wide surface is not “placing” anything.

Everything else

<div>, <span> and friends accept style and read it back, and it means nothing — they never render. There is no ResizeObserver or IntersectionObserver that fires, because no element has a layout box.

For text, use MSDF atlases or a 2D canvas. There is no text rendering from CSS.