Oh, I think I do understand how this is supposed to work and what you want to do. I am telling you, this will not work, even if your solution would do what you say it would do.
Think about it and try to answer this. Imagine it as a pitch and me being an investor: If your solution cannot prevent and cannot detect a bad actor, but only self heal by removing bad actors*, in what way is your solution any beneft to the regular account scrutiny of any platform - be it publishing or social media.
* and this self heal is not even on a platform level. Bad actors are not removed from the network, but only one line of the graph is blocked. And only by people that previously had a trust connection. And only if they bothered to log in and do that.
I do not think you grasp what a graph is. You basically draw naive daisy chains in your examples. And you rely on isolating the bad nodes, by their immediate neighbors. Completely ignoring the scope and size of a social network. The bad node will have thousands of trust nodes and billions of different paths to you. There is no single point of entry, where only node F would have to block bad actor A, so trusting user U has no longer a connection to A and thus "heal" the network. Your examples show single point of entry for the bad actors, but that is not how networks are. A will have connections to friends F1, F2, F..., F100. And your graph concept needs to have all of them to remove the trust, so no path to U will be available. Also consider shill network nodes S, that could encapsulate A. One could not cut the trust lines to A at all, only by cutting all of the shill nodes S as well - if you would even notice that S are also controlled by A, instead of being naive regular users. So you would need to resort to marking A as an evil node ... and that is just banning accounts with a lot of extra steps.
It gets worse and worse, the more I look into the technical details. That file format is not even needed for your trust graph. And it completely fails for text postings. That trust graph could be stored in the trust network servers. File formats would only be usefull, if you have people outside of the network. Like a seal of approval. Similar to how Microsoft tries to sign executeables.
Calculating paths is costly. Your Kevin Bacon example only works for node sizes of a few hundred. Like known actors. Should your idea be implemented, it needs to be able to calculate paths in a network with billions of nodes. Yes, billions. Facebook has billions of users. Imagine doing path calclulation in such a network.
Your verification by calculating a path is making an O(1) action of checking valid membership to an O(n) action, with n being the number of nodes in your graph.
Anyway, your solution will not work for social media, and it will not work for detecting ai authorship in video games. Those two scenarios are completely different. That you even mention text and claim you would sign the text in a file format shows, how out of touch with reality this concept is for social media.
Oh, and to clarify: I argued under the assumption that your solution would work on a technical level. I have doubts it would, but that is how I argued. I tried imagine a scenario where your solution would be adopted widely. Where people use it and it would do what it says on the tin. And even in that scenario, your concept falls apart when scrutinized by comparison to how the industry deals with the problems you are trying to solve.
Video game developers as a whole are not scammers. And the scams you will encounter are not the type of scam that relies on selling ai works as human made. They do not need to tell lies. So essentially you are trying to solve a non problem. Where scammers do tell lies, is when they claim, that their game is not stolen and does not contain viruses, does not steal your data and does not infect your computer. That is the type of scam you will encounter on Itch. A lot.