Trying to work out in advance why a macOS baseline will not match a Linux one, before I spend a day on it.
My assumption was device pixel ratio: the Mac renders at 2x, the runner at 1x, so the PNGs are different sizes and the comparison never had a chance. Is that the mechanism, and if so is there a setting that pins the ratio so one baseline could serve both?
Accepted answer
Open the PNG and check before you plan around it — it will be 1280x720, or whatever screenshots.viewport says. The baseline comparison works in CSS pixels, so a retina display does not double the file, and pinning the ratio would fix a problem you do not have.
What actually differs is the text. Glyph rasterisation, hinting and subpixel antialiasing are properties of the operating system, and on a normal page text is most of the pixels that carry information. Two machines rendering the identical DOM with the identical browser produce images that are visibly the same and numerically different, everywhere there is a letter.
That is why the platform is a path segment rather than a tolerance: you cannot pick a threshold that forgives every glyph edge and still notices a heading that moved four pixels. So the honest design is two baselines, one per platform, each compared strictly.
Have a different question? Ask it.