Opinion from an unexperienced person:
I'd try optically spacing letters like /H/O/n/o without reference to the original spacing, and then compare and see if you can draw any conclusions.

I think relationship to the counters is key for letterspacing, so a quick question to arise is what has happened to your counters in your drawing the other weights?

I use the same spacing for all the weights, normally applying a bit more space (≈+10) to the thiner weights and a a bit less (≈−10) to the bolder weights – that is because it helps to the boldness of the typeface: if spacing is added, the font looks lighter. Of course, you need to test if the spacing added or deleted is the ideal amount and some modifications must be applied to some glyphs.

As for the kerning, I start with the same pairs for all the weights, but often the bolder weights need more kerning pairs since the glyphs are closer, and some pairs of the thiner weights look a bit odd (excessive in some pairs, not sufficient in others). In short: it is a helpful starting point.

> I think relationship to the counters is key for letterspacing

It's interesting because for Mixteca I've had to narrow the letters of the Thin to make them look the same size (I guess otherwise the counters are too big), which made me want to reduce the letterspacing too. Conversely, the bold had to be widened, and that made me open up the letterspacing. This is a hard game ;)

I now create a Thin, Regular, and Extra Bold, which I space and kern manually, then interpolate the other weights.
The Ultra/Black may require its own treatment.

I think the reason is that notan plays more of a role in the middle weights, where the balance and rhythm of positive and negative is crucial, whereas in the extreme weights (which are more "display" anyway) the role of proximity between glyphs plays a greater role.

These are quite different spacing strategies. Helvetica, for instance, is tightly fitted at the extreme weights, but more open in the middle.

Thanks Nick, that's an interesting observation about Helvetica. Yet again it seems there is no 'right' way to do any of this stuff!

Be very careful when you add new kerning pairs or fine-tune existing ones in ready weights. Even if you have consistent spacing between your weights, small random differences in kerning will shift bigger portions of text.

Marcin, I don't understand. Do you mean it's best to kern new weights independently?

Sorry... I meant: do not change existing kerning independently.

1)
So, when you have, say, Thin and Black as frames for your design, with acceptable metrics and kerning, you can get every other weight by linear interpolation.

2)
But when you make bigger adjustments in design - and spacing - of the interpolated weight, the lines change into curves.
So, after next interpolation, "linear" letterspacing must be adjusted according to the curve.

As you can see, the more interpolations, the more similarity between the curve and lines. Spacing of DemiBold and DemiBlack in my example probably wouldn't need any changes.

And now, when having the structure of our family, we start making changes to the kerning independently, we'll get a wavy output, where every weight has it's own, random line length.

What Nick said. You have to check the spacing of the middle weight too. Not only because the relationships have changed but also for mechanical reasons.

Take a run-of-the-mill sans-serif for example.

The angles of the AKVXWkvwy will likely change from heaviest to lightest. It's especially relevant in glyphs with diagonals. Even if you've adjusted the spacing and kerning for the heaviest and lightest weights, the transition of the negative space between some glyphs is not always linear.

The way the heaviest L or T fits with other glyphs is different than the way the lightest L or T fits with other glyphs. When you tuck a fat L into an O, there's a sharp corner and some negative shapes to deal with. With an ultralight, there's no sharp corner, just a line. If you don(t give it an extra kerning tuck, the gap hay look too wide. In my experience, the transition between weights for these situations is rarely linear.

Ultralights have problems that medium and heavy weights don't have. Periods, commas and quotes are hard to discern in an ultralight. A bit more space can help make things clearer. But a medium weight doesn't really have the same problem. The periods, commas and quotes are quiet easy to discern. If you rely on a linear transition for the medium weight, the spacing of these glyphs may seems too loose.

I see. Thanks for your insight. I guess the answer is that the spacing of the regular might be a starting point, but that I should be prepared to space and kern it all over again.