The practice of underexposing to prevent any possible hilight from being clipped is not exposing to the right. Its exposing to the left. It will produce a dark raw image which is then lightened in post (increasing the noise). It does, as you say, give more hilight headroom.
Exposing to the right is deliberate overexposure, to move the midtones into lighter regios where the linear sensor count makes better use of the available bit depth. This results in a light raw image which must be darkened in post (decreasing the noise). It does, as you say, give even less hilight headroom.
I agree with the main point of the article; the available dynamic range of the sensor is insufficient. A given exposure will have lost information from the scene, and this will always be the case until we have non-linear sensors with greater bit depth.
I also agree that using RGB histograms derived from a white balanced image (or worse, a single luminance histogram; or even worse, a single green histogram) does not allow for accurate judgment of how well the 0 - 4095 range has been used for each channel and whether severe under exposure has happened on one of them, or whether one or more channels have clipped. But that also won't change until we have cameras with open source firmware or, at least, the ability to accept on-camera plugins.
So really, all that I am disagreeing with is that you labelled the 'underexposure' method ETTR - which is confusing, and the real ETTR method has different defects than the ones you describe.
Lastly, interesting site and keep up the good work!
The practice of underexposing to prevent any possible hilight from being clipped is not exposing to the right. Its exposing to the left. It will produce a dark raw image which is then lightened in post (increasing the noise). It does, as you say, give more hilight headroom.
Exposing to the right is deliberate overexposure, to move the midtones into lighter regios where the linear sensor count makes better use of the available bit depth. This results in a light raw image which must be darkened in post (decreasing the noise). It does, as you say, give even less hilight headroom.
I agree with the main point of the article; the available dynamic range of the sensor is insufficient. A given exposure will have lost information from the scene, and this will always be the case until we have non-linear sensors with greater bit depth.
I also agree that using RGB histograms derived from a white balanced image (or worse, a single luminance histogram; or even worse, a single green histogram) does not allow for accurate judgment of how well the 0 - 4095 range has been used for each channel and whether severe under exposure has happened on one of them, or whether one or more channels have clipped. But that also won't change until we have cameras with open source firmware or, at least, the ability to accept on-camera plugins.
So really, all that I am disagreeing with is that you labelled the 'underexposure' method ETTR - which is confusing, and the real ETTR method has different defects than the ones you describe.
Lastly, interesting site and keep up the good work!