The moat, ordered by what survives stronger frontier models

Assume frontier models become dramatically better at generating scenes, dialogue, plans, and game content. What remains difficult is the world around generation:

1. Kept world history. A World's attributed turns, information movement, relationships, artifacts, and consequences cannot be regenerated without becoming a different world.
2. Situated context. The accumulated machinery deciding what each playable entity can know, remember, access, and do is more defensible than a personality prompt.
3. Programmable Domain mechanics. Rules, tools, economies, clocks, and information regimes that share one authority and provenance model make new worlds cheaper without flattening their differences.
4. Parallel-world lineage. Scenarios, branches, and related Worlds create a library of controlled variation and emergent history rather than a shelf of static generated settings.
5. World and intelligence ecosystems. If creators, agent builders, communities, and rights holders eventually bring worlds and intelligences worth inhabiting, each becomes a reason for the others to participate.

The moat is not infinite generated content. It is coherent continuation under many wills.

The person's side

The five above are properties of the world. Their twin is the person's practice record — their own history of having been dropped into this world and read back (12). Kept history is what makes that record mean anything; the record is what makes the world worth re-entering. With a policy in the seat, the same kept history is what an agent cannot have memorised (13). Neither is a verdict on anyone.

The moat is coherent continuation under many wills — and what each of those wills carries away and brings back.