Why this architecture exists
Consequential AI systems do not fail only because models are weak. They also fail when coordination, authority, evidence and accountability are mixed or implicit. Astrynn organizes these problems into layers so one tool does not pretend to solve them all.
Three public levels
At the top, CABI 2036 is the research direction. At the first operational level, KARVYNX and QEVYNOR occupy distinct domains. Beneath them, ARIA, Evidence Fabric, ARIS, Astrynn Recon and Delivery Discipline contribute shared capabilities for evaluation, measurement, research, evidence and delivery.
Separation of functions
KARVYNX addresses coordination: context, dependencies, relationships among actors and systems, and agent architecture. QEVYNOR addresses institutional integrity and assurance: what basis can support consequential action and how uncertainty should be preserved when that basis changes or cannot be shown. Neither domain substitutes for the other.
Shared capabilities
ARIA contributes adversarial evaluation; Evidence Fabric supports continuity and utility of evidence; ARIS supports measurement and benchmarking; Astrynn Recon supports strategic intelligence; Delivery Discipline supports bounded implementation and integration. They support the system rather than multiply product brands.
What the map does not mean
The map does not claim that the whole architecture is implemented in production. It does not turn research into product, architecture into operational permission, or assurance into legal or technical authority. Public claims preserve those distinctions.
From research to application
Astrynn tries to move an idea from research to application only after it survives comparison, real-object evidence and clear boundaries. Implementation should be a consequence of evidence, not enthusiasm.