"Orange," "green," etc. in those graphs are just different models of lightbulbs. (Maybe using patterns - dotted, dashed, etc. lines - would have been clearer than overloading color.) For colors, look at the frequencies on the X-axis.
Anyway, the trick is that photoreceptor cells are sensitive to a range of frequencies, with weights centered on a particular color, and return a single floating-point output. So your blue cones cannot distinguish bright green and dim blue - that's only calculated by taking the input of the green cones into account, too. See https://i.stack.imgur.com/5snTb.png for what each cone's sensitivity looks like (in this graph, lines are light received by each cone, and the colors are semi-meaningful). If the goal is to not stimulate the blue cones at all, you need to emit as close to pure red light as possible. The wavelength of your light as measured by the single average or dominant color isn't as interesting as the spectrum of all light it emits and how much that overlaps with blue sensitivity.
One thing to point out is that the circadian rhythm is (largely) not directly influenced by the cones or rods (to our current knowledge). It is a third set of photoreceptors, the ipRGCs, which detect blue/green light and provide input to the circadian rhythm in the pineal gland/suprachiasmatic nucleus. The gray area is the sensitivity of the ipRGCs.
Also, that's a good point about the colors vs. dotted/dashed lines. I'll look into that, thanks!
We know the ipRGCs receive cone and rod inputs and that non-melanopsin responses (from rods and cones) drive a lot of the reaction at lower light levels.
> If the goal is to not stimulate the blue cones at all, you need to emit as close to pure red light as possible.
Ah, I gather this is why you use red lights at night when you don’t want to disturb your night vision as opposed to a light with any green component (yellow, orange, etc).
Anyway, the trick is that photoreceptor cells are sensitive to a range of frequencies, with weights centered on a particular color, and return a single floating-point output. So your blue cones cannot distinguish bright green and dim blue - that's only calculated by taking the input of the green cones into account, too. See https://i.stack.imgur.com/5snTb.png for what each cone's sensitivity looks like (in this graph, lines are light received by each cone, and the colors are semi-meaningful). If the goal is to not stimulate the blue cones at all, you need to emit as close to pure red light as possible. The wavelength of your light as measured by the single average or dominant color isn't as interesting as the spectrum of all light it emits and how much that overlaps with blue sensitivity.