Ok, so let’s just linearize pixelBrightness, then for simplification you could completely skip the EV part, since the f-number will be probably fixed and everything else is linear anyway, and we finally get: absoluteBrightness = pixelBrightness / ISO / expTimeSeconds. Using some unknown units ;-) so then calibrate to match the readings of your luxmeter, and that’s all!

looking at the formula it is not clear if your pixelBrightness is linear - it should be! (apply the inverse gamma correction if your camera delivers sRGB image)

if you are confused about the meaning of "linear", it is when values increase by the same factor (multiplying), e.g. 1/100 sec ISO 100 is equivalent to 1/200 sec ISO 200 which is equivalent to 1/1000 sec ISO 1000 and so on. you might have heard about old film speed rating DIN, used before ISO was introduced - and it was using logarithmic scale: for example doubling the linear ISO value is equivalent to increasing logarithmic DIN value by 3, similarly how the logarithmic EV value increases by 1.

correction: iso IS is linear and the EV is not, that's why there is nonlinear relation between the two

I'm voting to close this question as off-topic because it's about geometry and machine vision

yet the example shows that the picture becomes darker at 300mm, which is the most common behavior. anyway, this is irrelevant, i said zooming normally changes the brightness (which agrees with what the OP discovered from theoretical drawing and is intuitively understandable) unless something is changed to compensate. increasing the pupil is compensation. i can't even see anything we could disagree about ;-)

not arguing, just extending, the question mentioned a phone so i thought it would be nice to add something about smartphone raws

Now, while you can’t do much with such ridiculous RAW file, mobile image processing software is able to accomplish amazing results by merging data from multiple exposures, multiple lenses - or both. Of course it uses a lot of raw data, but not simply a “raw file”.

Exactly, four times as many levels but in the bright side where they are mostly useless (that’s why sRGB uses gamma). And I’m not saying that having more data is bad, only that “traditional” raw files, as known from regular photography, are not worth the effort in mobile (at least for the current smartphones).

and since we specifically asked about mobile, typical camera RAW files have 12 or 14 bits per pixel, while smartphone RAWs only have 10 bits, which does not even exceed the JPEG dynamic range and is only marginally usable. in short, don't bother!

contradiction again - if it receives 1/4 of the light it must be darker, if the lens is bigger to compensate, then it no longer receives 1/4 of the light. as i said in the previous comment, the constant f number gives you the intended behavior but it's not something that happens automatically. someone has to switch to a bigger lens when changing the focal length or you need a specifically designed zoom lens to keep the brightness constant. the natural behavior is darkening, the compensation is desired but artificial and not always possible.

"sensor reproduces the image (scene?)" is false in general. the sensor can only see what is being projected by the lens. in our configuration the lens receives the same amount of light per scene area, regardless of the focal length (apparently it's a constant aperture lens). this means, the longer focal length only small portion of the light will be spread on the sensor = darker image. you might assume that the sensor reproduces the scene illumination when a constant f-ratio lens is used. this is, however, not the situation described in the question.

wrong. "aperture" is not the same as "f-stop". zooming with constant aperture always darkens the image. of course we normally use f-stops for convenience, but the aperture, arising from the physical lens properties is more fundamental (especially in the context of zooming - the front lens is not going to grow wider to compensate for the increasing focal length)