This capability is no longer working properly in LR 5, 6, or CC 2015 and uses a fixed 5x5 pixel sample area regardless of 'Scale' or Zoom View settings. Please see the following test images for clarification. Tested with Windows 7 SP1.
It still can cause confusion and should be corrected to show averaged RGB values for the actual Scale setting sample area.
Here's another example that better illustrates the issue, with the WB eyedropper is set to maximum (17x17 pixels). The central cross-hair is placed on the corner of a 5x5 pixel black square (0, 0, 0 RGB) with a white background (100,100,100).
If the display reads only a single pixel then it should display 0,0,0.
If it is averaging over a 17x17 area then it should display (17x17) - (5x5) / (17x17) = 91.4, 91.4, 91.4.
Ok, you're right, the calculations from the above illustration would be:
So the result is a 5x5 average for the % both under the WB dropper area and under the histogram.
But the WB is still dependent on the full area, right?
And in the lower drawing in the first message, where it appears to be the central pixel is the value of the percentages, it is really that the zoom level of 1:2 is complicating things although it's unclear whether the zooming out is causing the eyedropper do use the non-zoomed pixels a couple out from the black so is in an all white 5x5 area at the original 1:1 zoom or something else, like using the central pixel is happening.
Re. 5x5 Area for sampling: behavior hasn't changed on my system between Lightroom version 4.4 and CC2015.6.1. The active white balance sample area for percentages shown is limited to a 5x5 grid. The loupe was modified to show greater detail but does not affect sampled pixel percentage area size.
Re. Correct White Balance: Both Lightroom 4.4 and Lightroom CC2015.6.1 appear to be setting the WB sliders to the correct value.
Re. Incorrect RGB values shown: There is a difference in the RGB values shown for Lightroom 4.4 and Lightroom CC 2015.6.1. A bug has been filed for this.
My testing contradicts your first point about the WB sample area always being a 5x5 grid.
The PERCENTAGES are always computed from the 5x5 grid.
The WHITEBALANCE is computed from the full grid whatever it is set to.
This is demonstrated by test images over on what turned out to be a hijacked thread about something else so the discussion continued over here on the feedback forum:
You can clearly see the sampled WB is warmer with the 5x5 and cooler with the 17x17 because the larger sample area includes some yellow.
Do you have any demonstration images to show your conclusion? It's possible it works differently in different computers.
I agree the sample area for the number display "in the tool" is a 5x5 grid no matter what size the grid is but the WB set when you click is set from the 17X17 (or other) size grid. You can easily test this by setting the tool to 5x5 and set the WB, do a CTRL-Z and then set the tool to 17x17 and set the WB. (the tool is NOT moved}. The WB numbers for the two sizes in the Develop Panel will be different if your sample area has a lot of changes.
If this is by design it is very unintuitive. Beings Lr is reading the pixels from the larger tool to set the WB it should be easy to reflect this is the numbers the WB tool is displaying when the size is changed.
On the Mac OS (El Capitan) this behavior is still the same as it was in LR 4.4. It appears that this behavior has not changed. If you can demonstrate otherwise, I would need your OS version and the Lightroom version where the Loupe tool is using the greater than 5x5 area for calculating percentages so I can verify. Can either of you provide that? If we can show it is a regression in behavior, I can file it as a bug. Otherwise, we are looking at making more of a feature request to get the changes implemented.
1) The Eyedropper tool's RGB values correctly show the "average value" for the selected sample size (5x5 through 17x17 Sample area) and NOT a fixed 5x5 sample reading. Good!
2) The Histogram RGB readings display the "average value" for a fixed 5x5 sample area. This is the same behavior as in LR 6.10. The Histogram's RGB values may differ from the Eyedropper tool's RGB readings .To avoid confusion it would be better if the Histogram displayed the same RGB readings as the WB Eyedropper.
3) The actual applied WB settings are correctly calculated using the Eyedropper Tool's sample area as selected in the Scale slider (5x5 through 17x17 selection). Good!