CSS coverage

Scope here is chosen so that what is claimed is browser-verified rather than approximate: every supported feature is asserted against geometry extracted from a pinned Chrome — 5,304 conformance fixtures at 0.1px tolerance, each in four variants (border-box and content-box, LTR and RTL). How correctness is verified describes the pipeline.

Implemented

Three layout modes, composing freely in one tree:

• Flexbox — both axes and both reverse directions, wrapping including wrap-reverse, grow/shrink resolution, flexBasis, baseline alignment, and the automatic minimum size.
• CSS Grid — explicit templates with fixed, percentage, auto, intrinsic (min-content / max-content), flexible (fr), and fit-content() tracks; minmax(); repeat() with counts, 'auto-fill', and 'auto-fit'; implicit tracks via gridAutoRows / gridAutoColumns; auto-placement in all four gridAutoFlow modes; line- and span-based placement with negative line numbers.
• Block layout — normal flow with margin collapsing per CSS 2.2, including collapse-through, and the legacy text-align inheritance for block children.

Across all modes, the full box model:

• min/max constraints and the automatic minimum size
• aspectRatio, including its interaction with min/max transfer
• percentages (written '50%', stored as fractions), resolved per CSS — against the containing block's width on all four sides for padding, margin, and border
• auto margins, absolute positioning with inset stretching, RTL via direction, both boxSizing modes
• gaps, alignment including the safe variants, scrollbar gutters, and overflow's layout effects (scroll containers get a zero automatic minimum)
• measure callbacks for content-sized leaves
• browser-faithful pixel rounding (cumulative, seam-free), or exact unrounded output

Not implemented

Deliberate scope boundaries, not roadmap gaps — each would either grow the library toward being a framework or dilute the browser-verified claim:

• Rendering, styling cascade, or a DOM. The engine computes geometry from resolved style data. No paint, no selectors, no inheritance.
• CSS string parsing. Styles are structured data ({ fr: 1 }, { min, max }), not '1fr' or '10px 20px' strings. The single string form the library accepts is a percentage ('50%'), which is a spelling rather than a parser — one number, one %, no calc(). Every other CSS-ish string in the docs playground is parsed by the docs site, never by the library.
• Floats and inline layout. Boxes only; text lives behind measure callbacks.
• calc(). Values are numbers by the time they reach the engine — compute arithmetic in TypeScript, where it is ordinary code.
• Named grid lines, grid-template-areas, and subgrid. Grid placement is by line number and span.
• The order property. Layout order is the child list; reorder with insertChild.

Known divergences from Chrome

The policy: any engine-vs-Chrome disagreement is either fixed or documented in KNOWN_DIVERGENCES.md with a minimized fixture and a spec citation — zero silent divergence. Deliberate divergences are currently one class-level entry and no fixture-level ones:

• Cyclic percentage resolution. A percentage sized against a containing block that is itself sized by its contents (max-content sizing) is loosely specified (css-sizing-3 §5.2, "behaves as auto" in cyclic cases). Chrome effectively re-resolves such percentages after layout; this engine resolves them as indefinite during intrinsic sizing and does not re-run layout. Matching Chrome would require multi-pass relayout. In practice: give the containing block a definite size and percentages resolve identically to Chrome.

Defaults differ from CSS

The style defaults are the engine's own: display: 'flex' (CSS: block) and boxSizing: 'border-box' (CSS: content-box) are the two that change geometry relative to a stylesheet author's expectations. See Style properties.