A new exposure triangle for raw shooters

Here's a graphic that displays the light levels recordable at various ISOs in a camera with a base ISO of 100. It ignores read noise.

Comments?

Seems somewhat plausible for a camera / sensor with an ~12 bit linear ADC.

However, I don't see how it accounts for model specific highlight headroom.

And afaik there exist cameras / sensors with various bit depths, even dual output gain designs (e.g. Arri, Panasonic).

Or 13, 14, or 16 bits of precision.

That's a myth, measured with respect to full scale. I've only tested one camera with significant highlight nonlinearity before clipping, and that camera only did it at one ISO setting. I think the myth got started because of the different ways that camera's metering systems are set up.

As I said in the OP, I'm ignoring read noise here. And precision doesn't play a part in this graphic if the ADC precision is12 bits or more, so I'm not understanding what you're getting at here.

I wonder if that myth was started by including highlight reconstruction, i.e., the camera’s behavior after clipping of at least one channel. I have observed different amount of highlight reconstruction potential with different cameras (Leica M), but have not looked into it in more details. Highlight recovery is a game of luck as one never knows whether color will be preserved or not. Nonetheless, it is very helpful when exposing incorrectly.

As I understand it, highlight reconstruction is affected in the camera design mainly by the compromise matrix, and thus the CFA. But highlight reconstruction is not performed in camera, but by the raw developer.

Does photo diode-specific highlight headroom make sense?

Maybe ...

Shoulder is what I see ;)
Even more interesting in photo-voltaic mode (used in some technical sensor applications, I use one of those to shoot welding).

Any highlight headroom comparison and analysis that I have seen, evaluated how clipped raw files are handled by raw developer :(.

Sorry, I just don't buy that.

Can you elaborate? Jim's graphic is not related to DR, as I understand it.

Maybe I simple have no idea what Jim refers to when he says "light levels recordable at various ISOs" in the OP.
Not related to DR? D'oh!

With one exception at one ISO setting, all the cameras I've tested are designed so that the ADC clips before the curve above bends over far enough to be photographically useful.

In order to relate the graphic to DR, I'd have to bring in read noise and photon noise, and also pick a SNR threshold. That would make it camera specific. I didn't want to do that. DR is full scale over mean at some threshold SNR, so you can't read DR off the graphic. What the graphic does show is the effect of ISO setting on full scale with full scale measured in lux-seconds of exposure.

For camera-specific DR vs ISO curves, see Bill Claff's PDR charts.

Same here.

Neither read nor photon noise, so we may assume a SPAD-like photon counting sensor.
No light = no photon = DN 0
One photon= First stop of light = DN 1
Two photons = Second stop of light = DN 2
...and so on...
Add more bits in the ADC / DN and get more stops of light. Why stop at 11??

With no noise to worry about threshold SNR is 0. Add more precision (bits) and record more stops of light.

FWIW: Double the ISO, still may keep same DR 😉
photonstophotos.net/Charts/PDR_HighResShotMode.htm#Olympus%20System%20OM-1(LND64)

"above arbitrary base".

So I choose a base, arbitrary e.g. 3 stops higher than Jim's, same sensor as him.
3 stops more light at starting column in that graph, so the sensor must clip 3 stops earlier above my arbitrary chosen base: but Jim's graph still says 11 (not 8)?

Either the base is not arbitrary,
or the "11" is pulled out of thin air.