Property by property, against a browser

The test suite answers whether the engine passes the specification’s own test pairs, with no browser involved anywhere. This page answers a different question, and the only one on this site that involves another engine: put this declaration on a page, and does our render look like Chrome’s? Every property in the spec universe gets an example, both engines draw it, and the pixels decide.

140

Render like Chrome

pixel-exact, or inside the anti-aliasing band

141

Render differently

at least one example does not match Chrome

Not implemented

the parser does not know the name

163

Nothing to judge

no example moves Chrome, or Chrome has no such property

Counts are over the 489 standard, unprefixed properties; the full universe of 817 is one checkbox away. A property counts as differing when any one of its examples fails to match Chrome. Reporting its best example instead would make box-shadow look perfect on the strength of a shadow with no blur.

Standard, unprefixed only

489 properties

What the verdicts mean

✓pixel-exact: Our rendering of the example is identical to Chrome's, pixel for pixel — no further apart than the example's own noise floor.
≈below the visible threshold: Differs from Chrome only inside the measured anti-aliasing band (maximum channel distance 3 of 255). Curves, rotation and gradients land here.
✕renders differently: The property takes effect, but our pixels are not Chrome's. The difference map on this page shows where.
✕parsed, no effect: The parser accepts the name, yet our rendering is byte-identical with and without the declaration while Chrome's is not. The declaration is being swallowed.
–not implemented: The engine's CSS parser does not know this property name, so the declaration is dropped where it is read.
?no example yet: No example we built changes what Chrome draws, so there is nothing to judge the engine against. That is our gap in the sample set as often as it is the property having no static effect, and we cannot tell the two apart by measurement — so both stay in this one bucket, with the number of candidates tried published beside it.
?no browser to compare against: Chrome does not implement this property, so no example can demonstrate it and there is no reference to judge against. Usually a draft that has not shipped anywhere.

How it was measured

Each example is rendered twice — Chrome 148.0.7778.96 and our engine — and compared. Before a comparison counts, Chrome has to react to the declaration at all: an example where the browser itself draws the same picture with and without the property proves nothing, so it is thrown out. The search tried up to 40 candidate values per property and the sensitivity floor was 40pixels, which is what “Chrome did not react” means here.

The atlas is a snapshot from its own run (2026-08-27), engine 9843a84. The engine has moved since: conformance publishes bbd75e5 today, so a verdict here may be behind the current build.

Loading…