Every word of prose here was written by a large language model. Not assisted, not polished — generated. If that's a dealbreaker, better to know now than four stories in.
What follows is the honest version of how, because "AI-generated" covers everything from one lazy prompt to a lot of deliberate work, and those produce very different things.
What a person writes
Three things, all by hand:
The bible. The setting's rules — how the collar works, what it can and can't do, who the cast are, what each of them can't stomach, and a style guide covering point of view, tense, pacing and the tone range. Every story is generated against it. Most of the work on this project is editing these files rather than reading the output.
The premise. Per story: a situation, the exact wording of the agreement being signed, who signs it, who holds any say over what happens, where the story should land on the tone scale, and a list of beats — the events, in order, with staging notes. This is the actual authorship. A premise is usually several hundred words and gets revised more than once before anything is generated.
The judgement about whether the result is any good, and what to change if it isn't.
What the models do
A pipeline runs each story through several passes, each a different model doing a different job:
- Drafting turns the premise and beats into prose. This is the pass that writes the story.
- Fact extraction reads the finished story and records what it established — who signed what, how it ended — so a later story in the same thread can stay consistent with it.
- A consistency review reads the story against the premise and the bible and reports contradictions: a rule that quietly grew past its own wording, a specified detail that drifted, a story that landed at a different intensity than intended. It also reports what the premise got wrong, which is how the premises improve.
- Tagging produces the index metadata and the one-line summaries.
Nothing is committed silently. A story that comes back too short to be a story fails the run rather than replacing a previous version.
What this means for what you're reading
Continuity is real but shallow. Stories that declare a relationship to each other share established facts, and the links at the bottom of each story show those threads. Two stories with no link between them are independent, and were generated without knowledge of each other.
Inconsistencies happen. The consistency pass catches a lot and not everything. If two stories disagree about a detail, the disagreement is a bug, not a mystery.
Nobody depicted is real. The setting, the technology, the cast and every event are invented.
Nothing here is a model's opinion about anything. It's fiction generated to a specification, in a setting with rules, by someone who found the rules more interesting than the alternative.