If I establish something as true, I expect AI-assisted writing to respect it…

But generation is probabilistic. Even when the right context exists, an AI-generated response can introduce information that conflicts with something the writer has already established. The problem is therefore not simply whether AI can generate good writing.

It is whether the writer can quickly recognize when generated writing has drifted from the story they are building, understand the conflict, and decide what should happen next.

That led to the concept for Canon Check. Canon Check introduces that verification layer.