There are two aspects of dynamic range:

1. The capacity of the sensor itself (100%)
2. The numeric capacity of the raw file (for 14 bits, a value up to 16,383)

When there are more photons reaching the sensor than it can record for a particular pixel, it fills up and can go no higher than 100%. For most cameras at base ISO, when a pixel reaches 100%, all of the corresponding bits in the raw file will have a value of 1 (numeric value 16,383) and no higher value can be recorded.

At less than 100% of the sensor’s capacity any of the bits for that pixel might be 0 or 1 and the corresponding numeric values will be between 0 and 16,383. For a totally dark pixel, the sensor will record 0% and all of the corresponding bits in the raw file will be 0 and the corresponding numeric value will be 0.

At base ISO, if the capacity of the sensor has not been exceeded, the capacity of the raw file will not be exceeded either. The full dynamic ranges of both the sensor and of the raw file are being used.

What happens if you then drop the exposure in half and double the ISO (gain)? The sensor will only see values from 0 to 50% but the raw file will still record numeric values of 0 through 16,383. The recorded image will look just as bright so you may not see any change in the image – yet. The full numerical dynamic range will have been used but only half of the sensor’s physical dynamic range.

Repeat this and the sensor will only see values from 0 to 25% but the raw file will still record numeric values of 0 through 16,383. You can keep going. What is happening is that, although you are using the full numeric dynamic range of the raw file, each time you cut the exposure in half you are also discarding half of the remaining dynamic range of the sensor.

By the time you reach ISO 6400 you will only be using 1/64th (1.56%) of the sensor’s physical dynamic range but the recorded image will look just as bright because the raw file still contains values from 0 to 16,383. What has been happening to the noise? Each time you doubled the ISO (gain) you also doubled the numeric value of noise in the raw file. Your signal to noise ratio has been cut in half at each step, except for a little more effort the camera might apply to noise reduction at higher ISO values. You probably started to see some noise at around ISO 800 or 1600 and it continued to become more apparent the higher you pushed it.

As soon as you go above the cameras's base ISO it becomes possible to exceed the raw file's numeric dynamic range without actually exceeding the sensor's physical dynamic range. The pixel will be recorded as blown out with a value of 16,383 but the sensor's photon limit may not have been reached.

When someone says they are using the camera’s full dynamic range by using ETRTR/EBTR, if they are not also using base ISO, they are referring only to the raw file’s numeric representation of the image. They are not accessing the full dynamic range of the sensor.

Thanks for The detailed explanation and serious thought on this oft debated subject. As you note, the usable dynamic range can be bounded by the sensor or A/D on the the high end and noise on the bottom, and we can choose to use that range however we think best. If we have a camera with wide dynamic range and a low or medium scene DR, then you have lots of DR to play with and can underexpose to give ourselves lots of room on the high end (to prevent blowing out highlights). On the other hand, if your particular camera has a lower DR at base ISO, and/or you need high ISO because of shutter speed & DOF requirements and low light, then you need to be more careful as to how you use the DR you have available - depends on the camera, the scene and the exposure requirements.

Very similar to audio recording. When DR was limited to 60 dB or so because of tape and LP noise, the recording engineer pushed the gain right up to the limit- very close to overload and distortion (and occasionally beyond). But with the advent of DAT tape, dbx encoding and digital audio, suddenly they had 90 dB to work with and could allow a bit more headroom to prevent overloading on peaks without audiable noise.

Rather than follow specific rules, we need to understand the limitations of our equipment and processing and learn to judge light and the DR of the scene we're photographing. Personally, I find the old zone system (and an occasional spot reading) still most useful for evaluating a scene.

Happy Holidays everyone. My guests are arriving, and I'm fixing myself a drink 😈

Yep, dbx compression is a classic example of dynamic range (2:1) compression, (and their expanders are often used to undo some of the derogatory effects of peak limiting, which compresses and limits dynamic range). Over-compression and peak limiting in broadcast has long been used to give the signal more "dial presence", especially by AM broadcast stations. Unfortunately, the result is a signal with limited DR. It's not unusual to see less than 20 dB of DR (if that), especially from pop music stations. Digital recording has allowed the record/playback chain to achieve realistic DR that approximates that of the human ear, but it hasn't addressed the issue for standard broadcast stations who are concerned with SNR, "dial presence", and (supposedly) over modulation, but what's surprising is how few stations even own a spectrum analyzer and have no method for actually measuring peak modulation or modulation index or harmonics.