I really like your idea.
Spoiler It's similar to how one would fix the old "invisible wall immersion-problem in finite open worlds. Just set a core hand-crafted territory in the midst of a much larger procedurally generated world and players can then wander as they please. However, a certain regionally dictated sourness of milk, so to speak, will tend to encourage a preference for the hand-crafted area without making it feel like being fenced in with invisible walls from "PLANET-X!"
I tend to agree that the way to go with something like this is having 2-3 or even 4-6 hand-crafted settlement populations attached to key settlements, each with it's own distinctive architecture and building style, in amongst the dozens or more of generic settlement possibilities and generic settlers who, incidentally, would have no interest in sharing their story with the "Sole Survivor" because most people would prefer saner friends than the "famous" and more trustworthy friends than the "powerful". What some players may not realise, however, is that there are some very well thought-out back-stories for the Longs (Sanctuary Hills). Given the fact that this is an RPG and not a novel, such details take a little finding and while we all know that their stories started somewhere before Concord, getting to where it started is key to uncovering what happened - which gives a view, through the eyes of the Longs, of what it was that caused a general loss of confidence in the Minutemen. Covenant also has a remarkable amount of depth to it. There are eight other places hosting designated settlers:
Spoiler Norghagen Beach,
Warwick Homestead,
Somerville Place,
Greentop Nursery,
Oberland Station,
Tenpines Bluff,
County Crossing and
Abernathy Farm.
So, to my eye, there seems to be the potential for some core hand-crafted stories right there - although I'm yet to get to "know" the denizens of those other settlements. However, I can say that we already have at least two core settlements that are deeply written and appropriately integrated according to the medium (i.e. presented as appropriate to an RPG as opposed to the kind of linear presentation appropriate to a novel).
Ideally, the optimum number of optional, generically-manned settlements would have to be enough to raise the total number of settlements to around 180 at the player's option (i.e. if you don't want the site as a settlement, don't assign settlers to it). If network strategy mechanics are to be implemented effectively, then this is roughly the best compromise between difficulty and what human beings can handle in terms of complexity. As I've pointed out before, with respect to games revolving around network strategy, the concept is thousands of years old (easily very much older than Chess) and the numbers were optimized for playability (i.e. complexity vs. nominal human performance) a very long time ago. Why is this important? Well, Fallout and The Elder Scrolls already get strongly criticised for lacking difficult elements or minigames and something like this is a magnificent opportunity to introduce some truly cerebral challenges to gameplay for those who want to be challenged and which would scale, intuitively, according to the innate complexity of any given player's style of gameplay.
Myself, I only really focus on 2-3 settlements while using most other settlements I can open up as resource-gathering outposts. Presently, my provisioning map is reminiscent of Rome, and every time I visit my answer to Rome, I keep wondering why every supermutant in Boston hasn't descended en masse on the place. At this end of the scale, I think it would be fantastic if it really did matter how settlements were connected (provisioners), especially if connecting a given settlement to one provisioning network or another, really had significant and far-reaching consequences that can be shown to make sense according to a set of discoverable and consistent processes with the kind of deep relationships seen in games like https://en.wikipedia.org/wiki/Go_%28game%29 (For a deeper treatment, see: http://senseis.xmp.net/). This kind of depth could be achieved by introducing very simple and easy (i.e. cheap to implement) game mechanics such as using the number of provisioner routes, both to and from a settlement, as a multiplier for things like
- the probability of attack on the corresponding settlement and
- the number of hostile NPCs involved in any given attack on the corresponding settlement
For example, it would make so much more sense if a settlement with 7 incoming provisioner routes and 5 outgoing provisioner routes (i.e. total connections = 12) was four times as likely to be attacked and, generally, faced four times as many hostile NPCs per attack as a settlement with only 1 incoming provisioner route and 2 outgoing provisioner routes (i.e. total connections = 3). It would also add the kind of consequences to gameplay that I think many of the "it's too easy" critics really want. And making these mechanics interesting would be as simple as taking other things into account. e.g. provisioner route length (the longer the route, the more attention it draws and the more risk to the provisioner). This resolves nicely to e very simple strategic game where the player is only operating 2-6 manned settlements, but would start to get rather interesting when settlement numbers got above 18-20 (for those players who really are craving for a challenge).