People keep asking why coding gained dependable feedback loops before writing did. It is easy to assume writing just needs bigger models and more parameters, but that explanation only goes so far. In real production settings, the bigger problem is usually weak evaluation signal. If a team cannot tell whether a rewrite still serves the intended story meaning, output volume increases while coherence degrades.
“Creative writing isn’t limited by data it’s limited by reward functions.”
— Aidan Clark (@aidan_clark), Feb 6, 2026
That line matters because it shifts the conversation away from raw capability and toward objective design. A model can generate elegant prose and still fail narratively if it is optimizing the wrong thing. Writing systems often get rewarded for local fluency while missing global structural drift. That is why teams feel like they are polishing constantly without stabilizing story intent.
“really, really doubt great writers have unique reward functions. hill-climbing feels antithetical to creativity. reward functions define which hills to climb. wrong tool for the job.”
— davinci / Leo (@leothecurious), Feb 7, 2026
This pushback is fair, and it should not be hand-waved. If the reward target is “sound like a preferred style,” optimization quickly becomes homogenization. Writers are right to reject that outcome. The useful move is not abandoning evaluation, but redefining what gets evaluated.
The practical split: intent versus telling
“With Dramatica we have an objective method for evaluating thematic intent (underlying subtext). Does not address the ‘telling’ which is more subjective/artistic in nature — but you can run evals against the ‘completeness’ and integrity of a story.”
— Jim Hull (@jameshull), Feb 2026
That distinction is the core architecture in Dramatica. A Storyform is a structural specification of narrative intent, and the verifier checks whether a draft remains aligned with that structure across Throughlines, Dynamics, and progression. It does not attempt to produce one universal score for literary style, humor, or emotional texture. Those choices remain human and contextual.
This separation helps teams in practice because it reduces category confusion during revision. Structural checks catch contradiction early, while editorial discussion can focus on voice and impact without pretending those choices are objective. Writers still decide how the story sounds, but they do so inside a clearer narrative frame. That usually means fewer circular notes and fewer late-stage rewrites.
A muse still has value because it helps writers begin and discover unexpected language. But a muse cannot reliably tell you whether draft seven still argues the same thing draft one argued. That job requires verification. When intent is instrumented and telling is protected, creativity becomes easier to preserve under pressure rather than harder.