A new exposure triangle for raw shooters

The 11 was necessary to deal with the range of ISOs.

BTW, for scenes wider than 11 stops lens flare starts to undercut linearity.
www.rawdigger.com/howtouse/Dynamic-range-fair-share-of-flare-and-glare

Oh, so you just choose the number of stops so they match the number of ISO ranges in your graph, thus getting an isosceles triangle.
Nice trick.

Thank you, but it's not an isosceles triangle. It's a right triangle. It's just a teaching aid.

This is camera-specific, but it does show how both ends of the SNR=0 EDR vary with ISO setting. I need to work a bit more on this presentation.

Right triangle or not, if row and colum labels have the same distance increment it's isoceles for sure.

But anyway, your (OP) pic resembles a staircase more than a triangle.

You could make it an isosceles triangle if you did that.

Pixels don't have to be square. 😉

What definition for ISO are you using? What does the camera do when the ISO setting is changed? Do all cameras do the same thing when the ISO setting is changed?

Whatever the camera manufacturer chooses.

In the vast majority of the cases, it applies conversion gain, analog gain, and/or digital gain to the sensor photodiode charge..

Not all. The Fuji GFX 50S and 50R are an exception at some ISO settings. But not the GFX 50S II.

For dual-gain sensors, the diagonal should have a step at some ISO. At least that's what photonstophotos shows for my Fuji gear.

But in general, I agree with the graph's sentiment.

Dual conversion gain sensors do not affect clipping. They affect only read noise.

True enough, but dynamic range is always in reference to a baseline noise level, right?

I do not follow. The graphics does not relate to DR.

Well, the two conversion gains should give different absolute clipping levels in the photosites, but it doesn't usually show because the digitization clips lower than FWC.

Most references to FWC that you see are really about the digitization clipping at base ISO, not the actual "well". It is impossible to measure FWC on most real world cameras, unless you can hack the system and use a lower analog gain after the source follower than the firmware normally uses.

I don't consider what happens at the photosites clipping. It's more gradual than that. The ADC clips hard.

Correct. That's what this graphic shows.

Here's a teaser on where I'm starting to take this:

Well, I stuck with the language that I replied to. I don't know what happens with increased input, but I would assume that the two different conversions gains had different asymptotes after they also went non-linear at different levels.

When you say different levels, do you mean charge levels or voltage levels?

Outside the pale of course, but the first Foveon (F7) said "Well capacity is approximately 77,000 electrons per photodiode but the usual operating point (for restricted non-linearity) corresponds to about 45,000 electrons"

kronometric.org/phot/sensor/fov/SD9%20sensor%20in%20depth.pdf

Photon levels.

Not all, as Jim just said.

On one of my cameras, the ISO setting does not change the raw data at all.

I was talking about the asymptote value, not the stimulus.

Then you have the other extreme, like the Canon G9 I have (CCD), which uses analog gain to vary all ISO settings, but there is no visible difference at all. The camera goes from ISO 80 to 1600, but if you expose for 6400 in raw mode at either 80 or 1600, the read noise looks exactly the same. I coined the term, "gain in vain" for this camera. It may as well just use the gain from ISO 80 at all ISOs. Yes, the histogram is very natural and filled at ISO 1600; it's not an arithmetic push. The pre-gain read noise is just so much higher than the post-gain read noise (even at ISO 80) that you can't see the post-gain read noise (which means it probably has little spatial correlation).