Seems like they have an entire speedrun in there.

TW: suicide

more context

Rc / Arc are the real CoW types (they have „methods“ for that use case).

Also (one of) the word(s) for husband is 主人 which is the Korean 주인 which means "owner" 🤨

Look into target features, here‘s an example from my code: github.com/LiveSplit/Li...

My language wasn't set to Japanese, so it's a bug regardless. Though going from 0 to 11 makes so much more sense mathematically, that I don't actually mind.

Apparently it‘s so bad, you can start a lesson without a sufficient amount of energy and no warning, so you have to quit midway through, discarding all the progress.

To highlight why it's the better choice than sRGB, I extended the comparison to use colors. sRGB completely fails at blending green text on top of the red background and has dark anti aliased regions, whereas the others handle it correctly, but fail at preserving the text weight (except my "Oktext")

You simply take Oklab and replace it's 3.0 gamma with 2.2 for text. It's the best of all worlds, as it preserves most of Oklab's advantages, but works well for text (weight should look the same as sRGB). From top to bottom: Oklab, sRGB, linear, "Oktext":

As it turns out Oklab is not suited for text as it also does not correctly preserve the weight of the text. sRGB is actually the best that we got. However, you will need to blend colors, so sRGB actually does an awful job. So I came up with "Oksrgb" / "Oktext" (?). (next post)

For text you have the huge problem that rendering it in linear space causes the text to have a weight issue. Black text on a white background will appear thinner than white text on a black background. So you want to do it in a perceptual color space instead. (next post)