Hi Si
I have refrained from commenting because I have not done the experiments you mentioned in your OP.
But I cannot see much point in forcing the R pixels to become activated in a scene whose DR is within the capability of the S pixels. The reason for the R pixels is to extend the highlight DR. The situation hos describes is surely what the R pixels are for.
Equally, I see little to be gained in trying to relate your R-pixel-using-exposure to the reading of the internal meter or an external light meter. The internal meter is giving an average value for the whole scene. What you need to be considering is spot metering the shadows and highlights separately and seeing what the total DR of the scene is. For less than a certain value (probably about 7 stops), you can fit everything into the S pixels.
I always use manual exposure anyway and only ever use spot metering. I simply ensure that the highlights are not more than 2.5 stops above mid grey if I want to retain any detail in them. 3 stops if I want them white and on the verge of blowing. 3.5 stops if I want them pure white. Any more than that and the falloff will be artefacted. I don't believe in "recovery".
The external meter is usually metering incident light, yes? And it is therefore telling you the exposure that will make an 18% grey card mid grey in that light. This is no help for the use of R pixels. If you only need to capture objects lit by incident light, the metered exposure will do that happily with S pixels alone, as far as highlights are concerned. The problems come when you actually include a light source (or specular highlights) in your scene. I mean, shooting a scene that includes the ball of the sun is a different matter from shooting under the incident light of the sun. Or, less extreme, shooting a scene that includes bright sky can also exceed the DR capability of S pixels.
You mention noise from the R pixels in sky that is only about 1 stop over mid grey. Yes, I would expect the R pixels to be noisy at that light level because they are about two stops less sensitive than the S pixels as well as whole lot smaller. But I am wondering if the fact that you see the noise is due to the way ACR blends the R and S pixels. I haven't consciously checked for that phenomenon in HU. But I would suggest that a raw converter that treats the R pixels sensibly should not be blending them in at all when the lightness is only a stop over mid grey. I think this problem is because ACR blends them in regardless of whether or not they are needed. Whereas they should be used only for highlight DR extension. The blend should be examining the signal level of the R and S pixels at an image location and its neighbouring locations and making an algorithmic decision about how to blend so as to get maximum DR, minimum noise and smoothest tonal gradient.
So I think that using a different raw converter from your favourite choice might well remove the need to look for exposures that do, or do not, fire those little R pixels and instead allow them to be used for their intended purpose of highlight extension.
I routinely use HU and I have it permanently at 400% DR. But I am fairly confident that it does not blend in the R pixels 1 stop over mid grey. Probably....
--
******************************************************
I have a home on pbase
http://www.pbase.com/claypaws/
If you have the time to look
******************************************************
