and please don't think the student is trying to tell the master here
pleeeze
I thought (unless things have changed in the last 6 or 7 years), the measurements for one 'exposure' of an image were the same size across that image's DR, which is why measurement accuracy falls off with underexposure
You are right. Analogies always carry the risk of misleading.
If you are
judging, that the exposure won't be enough (based on metering or your feeling), you decide for the "finer units" and that applies to all pixels.
how are we, as non-engineers, supposed to know which ISO settings are using hardware or software-based solutions?
We, non-engineers (and even those of us, who are engineers) are not supposed to know this. You don't find any specification from major camera makers stating this. The maximum is, that the ISO settings past the last real one are characterized as "expansion" or "high" and often they can be blocked from accidental selection; furthermore they are excluded from auto ISO selection. Likewise, the 1/3 stop ISOs can be blocked, which is comfortable for raw shooters.
Often there is an ultra-low ISO as well, for example 50; that too causes losing one stop from the DR, and that too is treated as "extra".
However, manufacturers generally are very tight-lipped about anything, which could lead to better understanding of the cameras. For example Nikon never explained in cleartext, what the lossy compression means; they stated only, that
in most cases it does not cause any image degradation. Canon do not confirm the method applied in the compression, even though it is known widely (all programs, which are processing Canon raw files on their own have to know the compression).
Perhaps this is self-generating: all feel, that they would get into a disadvantaged position by revealing some of the restrictions, because others are not revealing that and customers would make a wrong comparison.
This is happening already. Example: the Canon 1DsMkIII's highest ISO setting is 3200 (I do not know if that is real or fake, I have not seen any raw file yet). Nikon's D3 goes up to 25600. Most Nikon owners herald it as three stop advantage over the Canon, but the D3's 12800 and 25600 are fake. Sometimes I have to laugh reading the owners boosting about the greatness of ISO 25600.
Anyway, analysis of the raw files, often the histogram alone reveals these tricks. Followings are the fine histograms from the Canon 40D. Each red, green and blue column of pixels in the histogram represents a distinctive pixel level; gaps indicate, that there is no pixel with that level:
Canon 40D ISO 100
Canon 40D ISO 125
Canon 40D ISO 160
Canon 40D ISO 200
ISO 3200 (fake) vs 1600 is particularly interesting. The ISO 1600 pixel values get doubled, therefor half of the original pixel values occupy the entire numerical range of the pixel depth. Consequently, pixel values within the top EV of the dynamic range get clipped - without any gain in the shadows.
The following layered TIFF demostrates the effect:
Canon40D ISO 1600vs 3200. The file contans the exlanation of the layers.
The principle is the same for all cameras doing this trick. On the other hand, the "true gain" in details depends on the camera.
Now, compare the above with the way the Sinar back is operating: higher ISO does not affect the raw pixel data at all. See the histograms of the ISO 100 and 200 shots with the e54: the pixel values reflect the lower exposure with ISO 200. This way the "higher ISO" (which is none) does not induce clipping:
Sinar e54 ISO 100
Sinar e54 ISO 200
(The displayed shutter time is incorrect; Graham has explained, that the camera does not pass this info on the back.)