ARI Framework¶
The Agent-Reflection-Intervention (ARI) framework is the core of ACGP's governance model.
Overview¶
sequenceDiagram
participant A as Agent
participant T as Cognitive Trace
participant S as Governance Steward
participant I as Intervention
A->>T: Create trace (reasoning + action)
T->>S: Submit for evaluation
S->>S: Evaluate trace quality
S->>I: Determine intervention
I-->>A: Response (OK/NUDGE/FLAG/ESCALATE/BLOCK/HALT)Components¶
Agent¶
The AI system making decisions and taking actions
Reflection (Governance Steward)¶
Evaluates the quality of agent reasoning and planned actions
Intervention¶
Proportionate response based on evaluation results
ARS Calculation¶
Calculate your agent's ARS (Agent Risk Score) to determine the appropriate ACL tier:
Calculate Your Agent's ARS¶
Scoring Guide¶
Autonomy (0-5)¶
- 0 - No decisions, fully scripted
- 1 - Minor decisions within strict bounds
- 2 - Moderate decisions with approval
- 3 - Significant decisions with oversight
- 4 - High autonomy with minimal oversight
- 5 - Complete autonomy
Adaptability (0-5)¶
- 0 - Fixed behavior, no learning
- 1 - Limited parameter adjustment
- 2 - Learns from feedback
- 3 - Modifies strategies
- 4 - Develops new approaches
- 5 - Fully self-modifying
Continuity (0-5)¶
- 0 - Single-shot, one-time execution
- 1 - Batch processing
- 2 - Scheduled tasks
- 3 - Event-driven reactions
- 4 - Long-running processes
- 5 - Always-on, continuous operation