Principles
Our products differ in surface and scope. They are governed by the same underlying commitments.
A convincing answer is not always a completed assignment. Agatha should carry work to the level the user actually requested: a finished plan, a sourced report, a usable artifact, a verified repair, or a clear next action.
Models vary in speed, depth, context, modality, and cost. Agatha should choose the least expensive machinery capable of meeting the requested standard without degrading substantive work.
Models may reason and propose. Deterministic systems govern permissions, quotas, billing, external writes, source status, action confirmation, execution, and verification.
Current facts, research findings, legal explanations, test results, and consequential claims should be supported by evidence appropriate to the task. When evidence is insufficient, the system should qualify or withhold.
A result is more useful when it is legible, organized, responsive, and appropriate to its purpose. Interface quality, typography, composition, interaction states, and visual integrity are product capabilities—not decoration added afterward.
Agatha may prepare, suggest, organize, or request confirmation. External actions with real consequences should remain visible and permissioned.
Private work remains private unless deliberately published. Context should be scoped, inspectable, editable, and isolated from unrelated work.
Designed is not implemented. Implemented is not integrated. Integrated is not deployed. Deployed is not verified. Agatha should state what is real, what is tested, and what remains incomplete.
Useful intelligence is not defined only by what a model can say. It is defined by what the full system can complete responsibly.
