The new WebAssembly logo uses a colour, #654ff0, that is claimed to be complementary to the one used in a JavaScript logo, #f0db4f:
I've tried various online tools (#654ff0 gives #fd9815 or #ffd63b or #d9f04e) techniques like inverting the RGB values (#654ff0 gives #9ab00f), and inverting the H (#654ff0 gives #d8ed50) and HS (#654ff0 gives #e5efa0) of the HSV, but none of these are very close to #f0db4f. What am I missing?
These are pretty close to being complementary. The closest I get using the Adobe Colour wheel is #654FF0 and #F0DA4F - that's only one hex digit out for the green channel.
The difference between F0DA4F and F0DB4F is imperceptible to the human eye. Might just be down to a rounding error somewhere, perhaps even a typo.
Color theory is not a mathematical model. Color theory mainly talks about how colors on a physical palette are mixed, that's it. There is no agreement as to what the color wheel should look like. Certainly not a RGB wheel. In fact you can skew the wheel as you like it would yield different results but still be a color wheel.
You could also use non traditional primaries then you would get a wheel that's completely different. Color theory is more concerned about getting most of what you have got than absolute truth.
Now color science, which is way different from color theory tells us that: The opposite color to yellow is almost certainly blue. Why? Well inside your eye the chroma receptors are arranged as antagonistic pair of blue - yellow and red - green. So the signal our brain gets is colors on blue yellow axis and colors on red green axis. We still don't know how off axis is interpolated though but have a reasonable guess (but it keeps getting refined)
The reminder here is though that, none of the color operations you do are super meaningful unless your working in the Lab color model, or the polar version of it LCh and know the autobalance white. But still back to the color wheel, since the brain autobalances color, the wheel is valid even though it's overlaid with a color filter of somekind. But by now everything is so complicated that even rocket science is easier.
At the end of the day, different models yield different color wheels. Just because its not complementary on the wheel you look at does not mean there wouldnt be a wheel that does. As shown in other posts.
The definition of complementary color is "a color on the opposite side of a color wheel (geometry) that cancels the other making the resulting one neutral (achromatic, grayscale), black or white".
I added a couple of words in parenthesis. Not every color model is a wheel, a circle. You could represent them in a hexagon, rectangle, or triangle. In fact, they are not 2D shapes, but 3D solids.
And if you think of the "opposite" on a 3D solid, you do not get white or black, but grayscale. You can have cubes, cones, double cones, cylinders, spheres, or potato-shaped 3D models.
Take a look at this question: Is there a standard for color wheels?
Let us stick to color wheels, well, we do not have only one "color wheel". Yes, we have the more standardized RGB-CMY color wheel, and numerically you use that model when you use hexadecimal notation and operations.
But a color wheel that stays on an "artistic level" is RYB one. In this wheel, Yellow is opposite to Purple. It does not make sense mathematically, only perceptually. It is the standard color wheel used in painting.
My assumption is that the person doing that color scheme used something like: https://paletton.com/#uid=a1S4G0k++NyttZOLy+V+WuJ+5ob
And as you can see, that is more or less an RYB model.
