For anyone interested DPReview has a discussion on ISO here: https://www.dpreview.com/articles/9698391814/the-ins-and-outs-of-iso-what-is-iso. This is apparently part 1 of two - and the second part may turn out to be the more interesting.

There are a couple of interesting observations:
- ISO is essentially the product of the analog and digital signal processing applied by the camera to the sensor output; changing ISO changes the signal processing, not the sensor output (ie the sensor doesn't become more sensitive). This means that ISO is implemented differently by different manufacturers. I've seen other articles online which claim out that ISO values by one manufacturer are not the same as those of another - for example the tone curve for ISO 64 by one might be about the same as ISO100 for another.
- ISO is a measure of the lightness of the JPEG output of the camera, rather than the raw file; I presume this means that raw processors (particularly third party software) must therefore apply some predetermined tone curve (perhaps to emulate the manufacturer's JPEG tonal values?) to the raw file to produce an 'acceptable' range of tones when the file is imported.

Simon

The second post is now on DPReview: https://www.dpreview.com/articles/5426898916/ins-and-outs-of-iso-where-iso-gets-complex.

I don't pretend to understand all of this, but an obvious question is: if sensors are 'iso invariant' in that they can produce satisfactory tone curves across a range of lighting scenarios through digital signal processing, without resorting to increased analog signal amplification, why do we need ISO at all? We might need some in-camera algorithm to ensure we can have a useful image on the screen/electronic viewfinder (and to ensure we have useful JPEGs out of camera), and we might need some baseline indicator so raw processing software knows what to set as a starting point when the picture is imported (so you don't just have a black or near-black computer screen) - but beyond that? We could just as easily use a simple brightness scale rather than the old film-derived numbers of 100, 200 etc.

However, I expect I am missing something obvious.

Thanks Simon. I haven't read the article yet but have read in this area in the past and prepared an article for the newsletter some years ago on ISO and noise. In terms of why ISO is needed at all if we have ISO invariant sensors, there are a couple of things I can think of:

1. ISO invariant really means almost ISO invariant or close enough to that state that the sensor can be considered such. The the extent they are not, noise will still come from camera sources and not just from the randomness of light itself (photon shot noise).
2. Perhaps more relevant, most cameras only record up to 14 bit raw files. That is, for each photoreceptor on the sensor, a 14 bit data item is created when a photo is taken. This enables 16,384 different tonal values to be recorded. If you had a camera with a fixed ISO (that is, not ISO adjustment), some photos are going to be recorded in a band at the top or bottom end of this tonal range. This will effectively reduce the data recorded for the image to only that part of the 14 bit tonal range the sensor is capable of recording to a small fraction - maybe 1,000 or a few hundred tonal values. The quality of such an image will be pretty awful as there isn't enough information to faithfully record a reasonable rendition of the scene photographed.

At least, that's how I think of this and understand it. I'll look forward to reading the article.