AI GM Context Discipline

StoryRig is built for an AI GM, so rule quality depends as much on where instructions live as on what they say.

Keep one short standing instruction: StoryRig Core. It contains rules needed across almost every scene—player authorship, when to roll, the outcome bands, mixed-hit discipline, and canonical lookup.

Situational procedures should not compete with it every turn. Position/Effect guidance, Threat procedure, Progress Track procedure, and Conditions/Growth belong in Partial or otherwise demand-loaded context unless your particular world uses them constantly.

Do not pin the library

Moves, Conditions, Threats, Equipment, detailed locations, and deep lore should be found when relevant. The GM does not need every Move definition in memory during a quiet scene with no rolls.

Canonical owners

Give each rule one owner:

• Move: trigger and outcomes.
• Condition: mechanical scope and recovery.
• Equipment: edge and limits.
• Threat: agenda and pressure state.
• Progress Track: advance, regression, and completion conditions.
• System Profile: stat roster and consequence vocabulary.

When a capability matters and its owner is not in context, the GM should read that record before defining what is possible. Do not rely on the semantic feel of a name. This prevents a model from turning “Influence” into mind control or extending a narrow piece of gear into a universal tool.

Check is not write

Write state only when play establishes a durable change. Considering a change is not itself a write.

Test the negative space

For focused AI-GM QA, define must change, may change, and must not change. A narratively good response that writes unrelated durable state still fails the test.