Another conclusion I disagree with is the author's insistence on only using integer scaling multipliers. The article talks about subpixel rendering and then mostly ignores that for the rest of the article, instead assuming pixels are all small black squares (which it isn't - "pixel density" isn't a particularly meaningful term when you're dealing with clarity on modern screens). IIRC, macOS renders text on a higher-resolution internal framebuffer and scales down, anyway; in practice I don't notice any difference in clarity on my MacBook screen whether running at an integer or non-integer scale of the native resolution.