Skip to content

Research & Evals

Dramatica gives writing what coding already has: a verifier. We do not optimize for a single scalar notion of “good writing.” We define narrative intent as a Storyform specification and measure whether drafts align with that intended meaning.

Use this section to review the technical framing and benchmark-style evaluations.

Start here

Three-layer eval model

  1. Constraints (compiler errors): does the candidate violate the intended Storyform?
  2. Alignment (unit tests): does the candidate express the intended Throughline pressures and Dynamics?
  3. Quality and diversity (search): within valid candidates, can we select among distinct strong options?

NOTE

The verifier narrows outputs to meaning-correct candidates. Human creators, editors, or separate style graders choose voice and taste.