Theory: Start a sketch (wherever makes sense - at asm level, skeleton part, in envelope, whatever methodology works for you and your stye) Draw the arc along the sprocket (always constant length, right?) tangent vertical line then tangent horizontal line (and maybe not tangent - note item about pitch adjustments around sprocket later in post). Add length dims to all three (to measure arc length, pick the ends of the arc and the arc itself and then plant the dim). Create an equation - horizontal leg = total distance - arc length - vertical leg . This keeps all three in sync after a rebuild, assuming that the vertical leg length drives it. If I have it backward, just change the equation accordingly so the horizontal length drives it. You can even use in context between component and vertical leg and turn that dim into a driven and the equation will correctly evaluate for the horizontal leg (ignore the warning about using driven dims - its just a heads up, not a limitation)
If I have the mechanism all wrong in what I described above, just ignore what follows... but let's pretend for now that I didn't screw up yet
You could then use equations referencing the segment dims to automatically drive your three patterns (instances and distance) but what fun would that be? OK, it's a little fun, but still - as you mentioned, its not going to account for when the pitch doesn't exactly line up, which might be important for interference detection, etc. I hate imprecision.
Lets go for the fence.
So now, edit the sketch again and add a single fitspline to the three segments. You have just turned three segments into one segment driven by the three (can do it in a second sketch if you like - again, personal style rules).
Now you *might* be able to use this single spline-segment for a curve driven pattern, but I'm not set up to try it right now and I kinda doubt it will work especially for the partial pitch situations. And, I'm pretty sure that curve driven patterns only work in parts so you have to pattern chain link bodies in a part. Or pattern features, then mate a single link component to the seed feature and use a feature driven pattern to make the assembly pattern (and, even so, I haven't tested if feature driven patterns are supported when the pattern is a curve-driven pattern - always keep an eye out for little pitfalls like this)
So how could we get this thing to update correctly and even incorporate the pitch? The following is a little Vegas (a gamble), but if you make small changes when you make changes I bet it will work. it might even work for big changes depending on how the sketcher feels in this release. In a new sketch, draw 20 line segments end to end (make sure no automatic relations are applied, like collinear, vert, or horiz) . Window select them and make them equal, then dim one to your chain pitch. Now you have a sketch that acts like a length of chain. Flop segments around to test it. That's kinda fun.
I *think* if you then make all the 21 pitch endpoints coincident to the fitspline, and one end coincident to the driven end of the fitspline, you will have a sketch mimicking a chain that will follow your layout sketch exactly even as the ends are altered. Note - don't constrain the other end of the chain, because that is NOT accurate - a faceted radius (pitch around a sprocket) is not the same as the arc length of that sprocket - there is an adjustment that needs to be accounted for with the way the sketch is set up with the tangencies and all.
Then you can mate chain segments to those 20 sketch lines. If SWx sketcher were really robust (and who knows in this case?) it would work and update wonderfully. I advocate that you test it first with just the sketch before burning time making mates. That would be fun to try out. Those pitch segments might follow the spline perfectly like so many ants on a scent trail. And if it doesn't work the failure will be interesting and educational. Sorry I don't have the time right now to give it a shot myself.
Ed
.