Your observation has helped clarify some of the structural issues I had identified in one of my own projects, so thanks very much for that. It's a limited scope open world, and I was going nuts trying to superimpose a series of gated puzzles on it for the sake of the narrative. My instincts kept telling me this was not the right direction, which is helpful but not much use for charting a new course.
What's funny is that my other WIP is also set in a limited scope open world, but since I cast it as a multiplayer game, it came pre-structured as a Capture The Flag or King of the Hill scenario. So there I have a thousand problems, nearly all technical, but I am not at all worried about incorporating player choice or interactivity, because the game will be obviously terrible if I don't. Obvious terrible is the best kind IMO because I know I have to fix it.
Ha ha, good point. And glad I could be of some small help to someone who has done a lot of work from which I have personally benefited! 8)
My CS skills are hobbyist-level, but the angle I've been working is to use a backward-chaining pseudo-AI, specifically RAPPER (T3/I7), to create autonomous situations that try to pull themselves into instantiation.
In other words, say you have a collection of Polti Situations. These are dramatic situations, like "Man Betrays His Wife For The Purpose of Bigamy With an Other Woman." There's 36 basic types, which are further broken down to one or two hundred fairly specific situations.
Let's say you have about two dozen of these. Each role in a given situation is a field that can be satisfied by a character. To qualify, there are rules attached to the roles. (X must be the spouse of Y.) Further, there are rules defining something like "betrayal," which would include different kinds of betrayal.
Every time a random event happens, one of these situations can tilt the randomness. The more activated the situation is -- the more rules are fulfilled -- the more power it has to tilt (pseudo)random events and thus pull itself into instantiation.
Frank and Betty have some nonzero chance of quarreling, and a quarrel has a nonzero chance of getting out of hand, and Frank and Debra have some nonzero chance of falling for each other -- or, Frank for Debra when Debra is uninterested.
This tends to shape sandbox events gently into dramatic events. To further shape these into *narratives*, dramatic events are more likely to pull themselves into instantiation when they're likely to meaningfully resolve past dramatic events. One character betrays another: that gives "revenge" and "forgiveness" situations more power to tilt random events their way.
If you got far enough, you'd need to tune it so that the same old situations didn't always crop up. In my notion, you'd want it sensitive to the PC pushing NPCs in one direction or another, but yet truly nonrandom, so that a given walkthrough would always give identical transcripts.
Raw materials for this one were Polti situations, "Games People Play," by Eric Berne, and the classics of game theory as templates for character interaction. Also some reasonable thermostat-type modeling of human emotionality, a notch more interesting than one finds in the Sims.
The notion of such a game is something I find weirdly intriguing. Also I wanted the PC to be switchable, so mid-game you could change to any other character.
A realistic look at the CS required for this one -- well, as I say, I'm a hobbyist. But if anyone finds anything pirateable in these notions, pirate away.
Your plan is a little too complex for me. I have this philosophy about modelling surface events and not attempting to model what's behind those events like individual characters' psychologies too much, except the main character. When I say that I wish to cast long-term consequences into psychological terms, I mean for the protagonist. It's not impossible that I will do the same for NPCs, but if I did that, then that NPC would be the entire focus of that chapter of the story, like Galatea or whatnot. I don't want to build too many complex underlying mechanisms for supporitng characters, because then they become too numerous and difficult for an ordinary player to *predict*, and you lose the sense of specific agency. Something similar has been tried in various ways (Erasmatron, anyone?) -- I find the problem with it is that it is often indistinguishable from randomness because it's just too complicated.
That's just by of explaining why I have steered clear of assigning all sorts of state variables to NPCs — I assign state variables to narrative events, instead; I am trying to skim the surface and fake it. Smoke & mirrors, et cetera. 8)
This doesn't mean that experiments of your style are not still valuable, because (1) I might be entiely erroneous in my rejection and (2) I might not really understand some key differences that make your iteration of the concept work where it hasn't for me before.
EDIT P.S. I think perhaps one differennce here is that my definition of a sandbox game does not include narrative events that weren't predicted by the designer of the game. Some people's definition includes that, but I think that's way too ambitious -- as in I don't actually believe that it is truly possible to achieve story (events, yes; story, no) in any meaningful way in a fashion that wasn't predicted by the dev. So I think of sandbox games as merely a cohort of individually non-compulsory, pre-designed narratives discovered 'accidentally' by exploring a relatively open field of play. Track-switches between the different narratives are careully allowed so as to ensure they will still make narrative sense; nonsensical track-switches are blocked and suppressed by the dev's noticing their incongruity in advance; and in this way a set of narrative branches are designed for the player to discover and travel preferentially. It's not like LEGO -- it would be awesome if it were but I don't see it happening and it's too ambitious a goal for me personally. I don't even think games like RDR and GTA really accomplish what they say they accomplish, so IMO, it's too ambitious for them too, although they are still good games.