ISO for digital photographers in 2023

Citation required that most implementations do that, or any in fact, when not constrained by actual exposure settings - as that would make them severely broken implementations.

Tried it on my Nikons, Sonys, Fujis, and Leicas.

In Auto-ISO mode:
In M mode, EC changes ISO only (no other option).
In A mode, EC changes ISO but not shutter speed until an Auto-ISO bound has been reached.
In S mode, EC changes ISO but not aperture until an Auto-ISO bound has been reached.

Why do you consider the implementation broken? It seems to strive to optimize IQ.

@JimKasson too...

That claim might be premature outburst due to the readout error, but let's dig into it.

First indicator was, that the live histogram does not move in a linear fashion when I alter exposure. If I increase exposure, pixels should rather move right till they hit the wall of saturation. But no. The shape of the histogram changes in a way that more and more pixels are stacking before the saturation point before hitting saturation.

That motivated me to take few exposure stepped shots, and indeed, RawDigger confirmed what is happening. The linear exposure stepping does not follow the RAW histogram output. Doubling the exposure is not doubling the numbers before saturation. The more the exposure, the less pixels occupy the space where they should be on the graph. I.E.: fully saturated, but they aren't.

The camera is EOS M6 II, again.

Default in-camera JPEG processing introduces a nonlinear tone curve in addition to the curve for the color space chosen. Usually has a shoulder region.

Tried it on my Nikons, Sonys, Fujis, and Leicas.

In Auto-ISO mode:
In M mode, EC changes ISO only (no other option).
In A mode, EC changes ISO but not shutter speed until an Auto-ISO bound has been reached.
In S mode, EC changes ISO but not aperture until an Auto-ISO bound has been reached.

I don't, but 'until an Auto-ISO bound has been reached' is key here - setting the ISO bounds to the full range of the camera makes it work as expected - not sacrificing exposure unnecessarily while keeping negative EC useful to protect highlights.

I've not seen that.

At "low extended" ISO setting some cameras use part of non-linear part of the sensor response.

The Q2M is like that at ISO 100, and I don't think that is labeled extended. Could be wrong on that last point.

The spec does not mark it as extended, and I do not think it is.
ISO sensitivity range
Auto ISO: ISO 200 to ISO 100 000
Manual: ISO 100 to ISO 100 000

This is not correct. For a given scene, it's the camera's auto mode and the chosen EC (if any) that is what "locks in" the final image lightness, not setting a fixed ISO.

Put a camera in manual and you'll immediately see you can change the final image lightness without changing the ISO.

(emphasis mine)

Apologies, this is also not correct. If an increase in scene brightness results in the camera's auto mode choosing a correspondingly faster shutter speed, the exposure--amount of light striking the sensor per unit area, as you said--remains the same (which is consistent with @MichaelFryd's statement).

But there's definitetly something strange going on at ISO 100 with the Q2N.

The context of my comment is for when in an Auto mode as I was quoting Michael's comment.

Yes, you are totally correct and what I posted in that quote is not accurate. I was thinking of a different scenario at the same time and unfortunately mixed them together.

I have now warned my proof reader for the last time :-)

Yes, that is why it is excluded from Auto-ISO range.

I just remembered my discussion about ISO 50 (SL2/Q2) with @bclaff and the Leica support.
Leica:
The native ISO for both cameras is 100, while ISO 50 is created electronically. The final picture is then pushed digitally, and for that reason the clipping is much more visible later.
Bill:
I measured about 65000e- at ISO 50 and 31000e- at ISO 100.
To me this confirms base is ISO 50. You cannot fool photon noise.

Yes, that what I was thinking, but the shaping kicked me to beleive that it is not just the thing that I am witnessing.

Will attempt to do more controlled test and come back...

Auto or manual, setting a fixed ISO does not lock in "a set final image lightness for the sooc jpeg."

So sticking with auto, fix your ISO and try different ECs. You'll see your image lightness changes while the ISO remains the same.

As an event shooter, I'm constantly shooting moving subjects in dim light. I used the "M" mode method for years. Now, I take advantage of Sony's excellent "Minimum Shutter with Auto ISO" feature, setting Auto ISO's ceiling at 25,600 and exposure mode to aperture-priority. With moving subjects, I set Minimum Shutter to "fast" [1/(2x focal length)] to prevent motion blur. For still subjects (e.g. venue shots), I set it to "slow" [2/FL] and let IS take care of camera shake. There are also "standard", "slower" and "faster" settings. The great advantage is that I can swap primes or change the FL of my zoom, and the camera adjusts the shutter speed to match the FL used. It even compensates for the longer EFL of Crop Mode. I never need to think about or change shutter speed except for switching from "fast" to "slow" and back. This is a godsend for fast-paced work such as weddings or the corporate events that are my bread and butter.

For example, with my 135/1.8, shutter is automatically set to 1/320 and bumped up to 1/400 if I enable Crop Mode. OTOH, when doing room or table shots with my 18/2.8, I switch to "slow" and the camera chooses 1/10, bumping to 1/15 with Crop Mode.

Since I started working this way 1-2 years ago, I've had virtually NO images that visibly suffered from subject or camera movement. I retain direct control over DoF, and when ISO drifts up, I clean it up with DxO's DeepPRIME noise reduction. For this kind of work, this feature is incredibly liberating and empowering, and it's equally effective for less demanding conditions, such as my daytime scenic walks.

Sony is not the only maker to offer such a feature, but it's the best implementation I've encountered hands-on. My Panasonic GX9's version is more basic, offering only "standard" [1/EFL].

@JacquesCornell

I find Canon's implementation works very well also.