GBSN Research · Open research demo

AI Data Governance Layer

Explore how ADGL governs what knowledge may participate in computation, how admitted knowledge may be analyzed, and what consequences a governed result may produce.

Public package 0.4.3 · Toolkit semantics 0.5.3
Public research demo: Do not enter confidential, personal, proprietary, regulated, or production data. This static demo executes locally in your browser and does not call an external AI model or a GBSN backend.
Knowledge GovernanceWhat information may participate?
→
Analysis GovernanceWhat may be done with it?
→
Consequence GovernanceWhat may happen as a result?
Not run

Reference policy (YAML)

Read-only canonical fixture

        

        

          
Input (JSON)
Editable — use synthetic/public test data only

          
        

      


      

        
Knowledge—

        
Analysis—

        
Consequence—

        
Status—

        
Audit events—

      


      

        

          
Execution result
Client-side reference evaluator

          

        

        

          
Audit trace
Logical demo events

          

        

      


      

        
Expected fixture
Published reference expectation

        

      

    

  


  

    

      

        

        

          
Illustrative profile (YAML)
Non-normative

          

        

      

    

  


  

    

      

        
Candidate execution-integrity semantics

        
The published paper reports 25 normative conformance checks. Implementation and adversarial testing added 11 candidate integrity checks in the current public reference toolkit. The browser demo implements the most relevant fail-closed behavior for the interactive cases.

      

      

        
Upstream terminal propagation
If Knowledge Governance blocks or pauses a case, downstream Analysis and Consequence stages do not report successful outcomes.

        
AnalysisResult contract
Pipeline cases require a finite confidence value and an explicit uses_inferred_missing flag.

        
Explicit ACT authorization
A machine action does not self-authorize. The action must match an explicit case-scoped CapabilityGrant.

        
Value and template constraints
Action type/template fields and finite non-negative value limits are checked before authorization.

        
Human decision gates
DECIDE outcomes remain with a designated human role, and DECIDE → ACT requires approval before action authorization.

        
Expiry and strict grants
Expired grants fail closed and unknown CapabilityGrant fields are rejected by the static reference evaluator.

      

    

  


  

    

      
ADGL architecture

      
Knowledge Governance

      
Determines what information may participate in computation and in what evidentiary capacity. Reference semantics include source admission/exclusion, quarantine, authority, priority, provenance, lifecycle, isolation, and model/geographic routing.

      
Analysis Governance

      
Determines how admitted knowledge may be transformed, evaluated, combined, inferred from, and validated. The current reference cases exercise analytical method eligibility, minimum evidence, confidence thresholds, and human validation.

      
Consequence Governance

      
Determines what a governed result is permitted to cause: INFORM, DECIDE, or ACT. Action cases demonstrate human approval boundaries and case-scoped capability grants.

      
Audit + Provenance

      
Audit and provenance span the governance trajectory. This static demo produces logical browser-side event traces for explanation; the canonical Python toolkit contains the full research audit implementation.

      
Important boundary

      
This Space is an explanatory static reference demo, not a production enforcement system, legal-compliance certification service, or substitute for the canonical Python implementation and conformance suite.