In reply to Extremely well-spaced & kerned fonts?:
BTW, I assume that anybody who uses no kerns less than a given value should have all the kerns in a multiple of that value*.
I don’t agree with that assumption. Remember that kerns don’t create spacing by themselves, they correct the default spacing provided by the sidebearings.
People may choose not to provide small kerns for efficiency in the design process (speed or reviewing and correction, and easier management of a large set of data), for size reduction (still an issue if you have a very large character set, also there are the infamous limitations to the OpenType GPOS subtable size), and for the reason that tiny adjustments may lie below a certain visual threshold of perception (especially if the fonts are primarily intended for small sizes).
Think of a string in a guitar: if it’s hopelessly off tune, you’ll want to tune it, but if it’s just off tune a tiny bit, you yourself may not hear it, and the audience often won’t either. Similarly, if the default spacing provided by the sidebearings is “off tune” just a tiny bit, you may neglect a small-value kerning pair, but if it’s “more off tune”, then you’ll need to tune it, but the goal of that tuning is getting it “on tune” again and not of changing the pitch by a preconceived shift.
Another example: if a wall that’s supposed to be “straight” is off-balance by less than 5 degrees, then perhaps you won’t necessarily change it, but if it’s off-balance by more than 5 degrees, then you will want to fix it, yet it doesn’t mean that your corrections will always be in steps of 5 degrees.
I’d put it differently: if you assume that you don’t kern by less than 5 units in a 1000 UPM font, then it means that your tolerance of spacing errors is 0.5% of the point size, so in 12 pt type you agree that the acceptable deviation from the “ideal” spacing is less than 0.06pt. I.e. you agree that your spacing may be off by 1 unit or 2 units or 3 units or 4 units.
So to be consistant, you would also accept that your other kerning pairs may be “off by up to 5 units”, so for example, you could cluster all the kerning values which are different by less than 5 units. I.e. if you have a kerning pair of –41, a pair of –43 and another of –44, you might change them all to –43. But it doesn’t mean that you’d need to make them –45 or –40 (i.e. in steps of 5).