GovernanceRecord verifiedCC BY 4.0June 29, 2026
CGP: Cyryx Governance Protocol for Agentic AI Execution
A published technical report defining seven protocol control domains and a protocol-level conformance model for agentic AI execution.
Applied Research
Research for systems that must leave the lab.
Cyryx investigates the engineering and operating questions that appear when AI is expected to support real products, workflows, and decisions. The purpose is practical: better architectures, better evidence, and clearer limits.
A publication record is not current implementation, conformance, certification, or client-outcome evidence.
Next chapter
Evidence before presentation.
Published work begins with a precise question and keeps its method, limitations, and public record visible.
Published research
Public records, not presentation claims.
Released work is listed with its public identifier, source record, publication date, and license so readers can verify and cite it independently.
Research directions
Six questions behind one operating system.
Execution architecture
How AI, deterministic software, tools, state, and people coordinate across a real task.
Context intelligence
How systems select, structure, constrain, and attribute the information used for a decision.
Evaluation
How representative cases, human judgment, and system signals can support release and change decisions.
Authority & governance
How action boundaries, review, escalation, evidence, and ownership become part of system design.
Cost intelligence
How model, provider, infrastructure, and human effort can be interpreted at the workload level.
Human-system interaction
How interfaces communicate uncertainty, evidence, control, failure, and recovery to operators.
Research discipline
Evidence before publication.
Public claims, publication records, identifiers, control mappings, and maturity statements are withheld until the underlying evidence and release approval are complete.
- 01
Investigate
Frame a precise question from product or system work.
- 02
Prototype
Build the smallest instrumented environment that can produce useful evidence.
- 03
Evaluate
Test representative behavior, failure modes, limitations, and competing explanations.
- 04
Integrate
Feed relevant findings back into product, advisory, engineering, and operating decisions.
- 05
Publish selectively
Release public material only after its evidence, attribution, limitations, and approval are ready.
From research to practice