The second of these findings does not surprise me at all: it is quite common for the bolder weights of a type to have a slightly taller x-height than the lighter weights, as this allows compensation for horizontal reduction of interior counter space and makes the design of complex shapes such as the lowercase a easier. Similarly, the x-height of lowercase italic letters are often shorter than those of their roman companions in order to provide optical correction for the longer diagonal stems, which can appear taller than identical height straight stems.

The variation in the height of caps is surprising though.

Alignment is not as important as apparent size.
Relative vertical proportions of ascender, x-height and descender are also relevant, which may explain the variation in cap height.

If you think there is a problem with the dimensions mentioned here, it would be best to illustrate it with a specimen of “real” running text, rather than in the abstract.

Many well-established faces have x-heights which vary according to weight.
Gill Sans, for instance.
It’s a practice which I’ve implemented in many of my designs.

I was taught that the heights need to vary for optical reasons. The numbers you have provided look good to me.

In FF Ernestine* there's an increase of 4 units (Em = 1000) between each of Light, Regular, Demi and Bold, for each of: x-height, cap height, ascender height and descender depth. People seem to agree that lining up the baseline remains sacrosanct, even in a text face. In display typography lining things up is much more important.

* http://ernestinefont.com/

I agree with John that going the other way with the caps is peculiar, but maybe it's this: cap diacritics -which go over the em a good deal- might need more room the darker they get.

hhp

Perhaps this will explain what I mean by speaking of "bizarre logic":

http://www.sanskritweb.net/fontdocs/source_sans_pro_adobe.jpg

By the way, the horizontal dimensions feature the same bizarre logic.

For instance, the vertical stem of the capital "T" of the roman/normal/regular version of Source Sans Pro (i.e. SourceSansPro-Regular.otf) has a width of 83 units, whereas the vertical stems of the capital "H" have a width of 84 units each. (For comparison, the vertical stem of the lowercase "l" has a width of 82 units.)

For instance, the vertical stem of the capital "T" of the roman/normal/regular version of Source Sans Pro (i.e. SourceSansPro-Regular.otf) has a width of 83 units, whereas the vertical stems of the capital "H" have a width of 84 units each. (For comparison, the vertical stem of the lowercase "l" has a width of 82 units.)

Uli, are you bothered purely by the numerical inconsistency or are these differences noticeable (assuming it is possible to be objective, post-analysis) in, for example, running text at 12pt? John Hudson posted last year about his requirement for a script to harmonise stem widths in a typeface he was developing and I am curious about how detremental this would actually be from a purely visual perspective.

Uli,
As everybody said, this is completely normal for optical compensation.

It's also a side-effect of interpolating from only 2 masters (Regular and Black).
In bigger families, people usually use 3 masters (Thin, Regular and Black) so there is no difference in x-height, Ascender, etc. from the Thin to the Regular. But then again, they slowly increase from the Regular as it reach the Black weight.

Ideally, the vertical metrics will remain unchanged in the middle weights, and will only increase/decrease as they reach the thinnest and blackest extremes. But using lots and lots of masters for interpolation is not very practical, as each master increase the complexity and the chance of inconsistencies.

In the low res digital age, little inconsistencies like this had better add up to huge optical advantages. Huge. But it's probably a lot better than any free fonts we ever release.

Uli, again: that "inconsistency" is not a bug, it's a feature. Some people even make the "z" shorter than the other letters (because it's the only lc with large "border" horizontals). Like Adrian Frutiger. Professional. It is in fact impressive (and maybe disconcerting...) that a free font has such subtleties.

Fiona Ross said something cool during her presentation in Yerevan in June: "For things to look the same they have to be different."

hhp

The vertical proportions are based on Robert Slimbach’s guidance to me according to his ever-evolving theories of reading optics. The basic premise is that capital letters and letters with extenders appear larger the heavier their weight is. The differences you noted are subtle enough to be noticed in print and to my eyes, Robert’s guidance proves true: the various weights do seem better suited to each other than if they were consistent across the range of weights. In proofs that kept ascenders and descenders consistent with the regular weight, the extenders seemed a bit cramped in the ExtraLight weight and a bit long in the Black master. In low-resolution settings these small deviations are not present thanks to hinting instructions that keeps vertical dimension parameters consistent across a range of weights.

In proofs that kept ascenders and descenders consistent with the regular weight, the extenders seemed a bit cramped in the ExtraLight weight and a bit long in the Black master.

Interesting. But/and the x-height letters behave the other way?

BTW, I think that's the closest we've ever come to a post by Robert. :-)

hhp

Wit' all due respect to 'de old man in 'na mud brick... at Linotype I'd been taught, (I know now), to quantize the alignments. But 10 years later, sitting with a brilliant group of interns and juniors, one of 'm asked "..so if the [sans] O and H don't sit on the baseline together, why do the "L" and "I?", and we talked and talked...

20 years later, I still I know this theory, but this isn't all of it. This is a puzzle for developers:)

Robert Slimbach's theory on that was about the same when he promoted it to me some eight or nine years ago.

I was stubborn, however, and had pretty severe misgivings about varying the cap height across weights in Hypatia Sans. In the end I kept the cap height consistent across weights. I'm willing to believe that could have been a mistake, though I guess it is reassuring to see that it is still a contentious question!

On the side, those single-unit variations in stroke widths may be the result of interpolation in multiple master space.