Most page layout programs allow for exceptions in their H&J settings, but I doubt that you can program a font to write those exceptions to the dictionary. Also, it's doubtful that a font could be programmed to switch H&J on and off at will. If the letter combinations are really that important, make them ligatures and be done with it. It may not be your ideal solution, but it's one that will work.

AFAIK, the ligature feature ceases to work when tracked out.

This kind of behaviour in a layout application makes sense for Ligature features such as the traditional "fi", and for the original exegesis of Contextual Alternates as something for cursive scripts (IIRC Caflisch, released in 2001, was instrumental in establishing this concept).

Now it's been grandfathered, and we're stuck with it, even if we want to use "calt" for pseudo-randomisation, where it would be appropriate to keep the effect operative whatever the tracking.

There has been a "rand" feature defined for several years, but it is no use unless supported by Quark and InDesign, which it isn't.

Shame, I've designed a couple of "pseudo-random" faces, and it's something that really interests me.

Actually, not really "pseudo-random", but "pseudo-natural", or "non-repetitive".
That's because if the goal is to make digital type have the natural irregularities of letterpress printing or ink writing, randomly selecting the iteration of a character from a group of variant glyphs won't do that, as statistically there will be quite a few adjacent doublings.

Well, it's not about certain letter combinations, but about randomness. It's pseudo randomness, though. I picked up the examples from Thomas Phinney and Nick Shinn (http://www.typophile.com/node/19625).
By the way, the problem applies to ligatures as well, doesn't it?

Edit: I missed posting this earlier. I was in preview mode... well, hello Nick. :)
Regarding the ligature feature, Indesign (CS3 in my case) breaks the fi ligature at about +16.

[Hello Alex]
+16
That also varies with font size.

You could crank up "cpsp" to provide a generously letterspaced effect.
This feature is implemented by the "all caps" command, command-shift-k on Mac.
(If the user doesn't want it, the shift key and caps lock do not activate it.)
And of course, nobody ever letterspaces lower case...

I haven't been able to test whether that would work, as I don't put "cpsp" in my fonts.

http://typophile.com/node/52854

Thanks for the link Grrrben. As John Hudson stated at the end of that thread, negative tracking values have more tolerance before breaking the ligatures. So my workaround now is to increase the spacing within the font. This way it is possible to use negative values up to -44 (InDesign). It's a crutch, but at the moment I can live with that. It's a rather scrawly handdrawn font with capital letters only, so the wider standard spacing is not an issue.

Thanks to you as well, Nick. Though, I tend to use the all caps feature in Indesign to get alternative combinations more quickly.
Speaking of quick access and convenience, does anyone know, whether there is easier access to Opentype features, especially stylistic sets, implemented in InDesign CS5? I have to muddle through three flyout menus to switch them on or off in my CS3 version.

I'd like to revive this threat for I've experimented with position values in stylistic sets. For InDesign users, it would be possible to have a wider spacing by choosing stylistic set 17, which simply has the following code:

feature ss17 {
position @ALL <0 0 50 0>;
} ss17;

By further experiments with y values I came to some interesting results with dancing glyphs or uneven lines. Funny stuff to play with.
One irritating issue, though: when it comes to hyphenation, some stylistic sets with positional programmings cause a missing glyph right after a discretionary hyphen. “.notdef” is displayed then. I thought, I could at least leave out the visible shape in the .notdef slot and actually it hides (while still marked as missing glyph) in InDesign. But when printed to PDF, a quoteleft (‘) shows up after the hyphen. Could somebody figure out, what's wrong here. Maybe I'm missing an important modifier letter in my font or the like?

The Stylistic Set features are specifically described in the spec as GSUB features, not GPOS. This means that you can't predict what applications will do if you put GPOS lookups into these features, so it should be avoided.

Hmmm, well… what a pity! This would really extend possibilities for random looking handwriting. However, thanks for clarificaton, John.

You could write-up and propose a set of Stylistic GPOS features.