Just a little misunderstanding... when i talked about the last bit I meant the 16th bit.

In theory with 8 bits we should have 256 (2ˆ8) different values and with 16 bits, 65536 values (2ˆ16).

But for 32768 values (32769, as we saw previously) we need only 15 bits (2ˆ15).

Here is my missing bit the 16th one.

Is the 16 bits thing just a name but in reality we have only 15?

I would guess so as it was stated clearly that calculations are based on 0-32768.

Is there a reason behind it?

ok you have a point here.

So out of the last bit that in theory could give 32768 extra values, only one is actually used

But if the 16th bit was fully used then the maximum value would be 65535, and loosing the midpoint, or a 17th bit would have been necessary to get the value 65536... to get the midpoint back. Am I right?

About calculations:

HSB using B

100% gives 32768 reading with eyedropper

1% gives 164 (seems not correct, should be 328) in grayscale reading with eyedropper but  R G and B all at 326 (seems not correct, should be 328, note that file is grayscale 16 bits but values readable in RGB)

2% gives 328 (seems not correct, should be 656) in grayscale reading with eyedropper but  R G and B all at 652 (seems not correct, should be 656)

10% gives 1773 (seems not correct, should be 3277, but in any case 164x10 is 1640 so 10% is not 1%x10, so we don't have anything linear here as you suggested) in grayscale reading with eyedropper but  R G and B all at 3274 (seems not correct, should be 3277)

etc

99% gives 32355 (seems not correct, should be 32440) in grayscale reading with eyedropper but  R G and B all at 32442 (seems not correct, should be 32440)

This gets me quite confused