I have written about ETTR a lot, but I still don't have a single comprehensive piece.
Now only shortly.
1. Perfect ETTR is trial and error. The following technique does not ensure (and none other ensures), that the exposure will be perfect; it only helps judging if the exposure was perfect.
2. The in-camera histogram and the clipping indication are based on the preview embedded in the raw file. This preview is created just like any in-camera JPEG, using the respective settings.
3. In order to coax the camera into displaying a histogram and clipping indication, which are representative the raw image, one has to "neutralize" the camera's actions involved in the process of creating that JPEG image.
This means
a. turning off sharpening (0, the leftmost position),
b. turning off contrast adjustment (0, the middle position),
c. turning off saturation adjustment (0, the middle position),
d. turning off color tone adjustment (0, the middle position).
Of course, a,b,c and d can be set in a picture style.
e. selecting sRGB for color space (it is possible, that certain models would require AdobeRGB),
f. using a neutral white balance template, also called Uni-WB (as it results in unity coefficients). This is the greatest factor.
Such a setup results in an ungly, greenish, useless preview and thumbnail, except for judging the exposure. The method is not perfect, for there is no way with the present cameras to avoid demosaicing and the color space conversion, which involves the nonlinear mapping as well. However, it works reliably. I have been using this all the time with my 40D, and the flashing indication is within less than 1/3 stop from the raw exposure.
I have a WB template for the 20D and one for the 40D, downloadable for the 20D
and for the 40D
.
This template should be stored on each card as write-protected to avoid accidental deleting. One can switch between the template and any other WB method.
If anyone has or creates a Uni-WB, it can be verified by Rawnalyze: the "as shot" WB coefficients are displayed on the Histogram panel. They should be between 0.95 and 1.05.
Further, long-long discussions and explanations on this subject can be found at
here
and here
and here