Canon-aware AI context
Canon-aware AI context is structured story memory prepared for use with an AI model.
Instead of sending the model a vague instruction such as “remember everything about this novel,” Arc aims to provide relevant entities, relationships, evidence, and review state for the task at hand.
Why raw prompts are fragile
A raw prompt can be too long, too vague, or too disconnected from evidence. As a story grows, a model may:
- forget earlier details;
- merge similar characters;
- invent missing continuity;
- over-trust summaries;
- revise against the author’s intent.
Canon-aware context reduces that risk by narrowing what the model sees and by keeping the source of claims visible.
What context can include
A canon-aware context packet may include:
- relevant entities;
- aliases;
- relationship edges;
- accepted canon claims;
- unresolved review items;
- source excerpts;
- author notes;
- constraints for the current revision task.
The exact content depends on the product surface and the task.
Author control
Canon-aware AI is not the same as autonomous writing. The model can suggest, compare, summarize, or revise with better context, but the author still approves the result.
A good AI workflow should make it easier to say:
- “This suggestion is supported.”
- “This contradicts chapter five.”
- “This is plausible but not canon.”
- “This voice is wrong even if the facts are right.”
Preview note
The shape of AI-context features may change as Arc evolves. The core principle should remain stable: AI assistance should be grounded in reviewable canon and evidence.