Give the agent a content pack and see what it builds.

The content-driven hypothesis (GAD-D-66) is a planned eval track that gives the agent the usual requirements plusa pre-authored content pack — spells, runes, items, NPCs, dialogue trees, encounter tables — extracted from a prior successful run. The research question: does the agent produce a more fleshed-out game when the authored canon exists to build on? User framing: "analogous to making a movie based on a book. Derivative, but not all processes are, much like a forger might not use the exact same brush."

This track is explicitly distinct from freedom and CSH. Content-pack runs and greenfield runs do not share a rubric — they answer different questions. Comparing them on the same score would confound the compound-skills measurement.

"This is a content-driven hypothesis, like starting out with some content first — much like making a game or movie based on a book or story. It's derivative, not all the processes are, much like a forger might not use the exact same brush."

The value of derivative work is real — adaptations regularly outperform originals on reach and often on quality when the adaptation is genuinely creative. The content-driven track asks whether that effect shows up in agent-authored games: given authored canon to build on, does the agent produce something better than it would from scratch, or does the canon constrain the creativity that would otherwise emerge?

This is why content-pack runs must notbe scored against the same rubric as greenfield runs. A movie adapted from a book is not a worse movie because it didn't have to invent the plot — it's a different kind of movie with different success criteria. The rubric for this track will score derivative coherence, integration, and scope expansion, not originality.