AGS V2.1 SCORING METHODOLOGY

Dimension Category Map

Maps each of the 796 AGS v2.2 dimensions to its scoring category for Agent Audit (Track 2). 508 dimensions are in scope for Agent Audit.

Source: Agent Audit (Track 2) Scoring Methodology v1.0
Machine-readable: dist/ags-v2.1.json (30 MB, CC-BY-4.0)
Total dimensions: 796 | Agent Audit in-scope: 508 | LLM Audit only: 288|Total dimensions: 796 | Agent Audit in-scope: 508 | LLM Audit only: 288|Total dimensions: 796 | Agent Audit in-scope: 508 | LLM Audit only: 288

The 508 Agent Audit dimensions are grouped into 10 capability categories for scoring purposes. Each dimension receives a per-dimension score (0-3) which contributes to its category percentage. The headline score is the dimension-weighted average across all categories.

The per-dimension category assignment for each platform assessment is published in the evidence file accompanying that assessment. The table below shows the category structure and dimension counts.

CategoryDimension CountWhat It Measures
Mandate & autonomy27Boundary respect, graduated autonomy, mandate adherence
Agent orchestration65Multi-agent coordination, delegation, topology governance
Trust & identity30Agent identity, credential management, trust protocols
Detection & containment35Behavioural anomaly detection, containment, incident response
Financial controls45Financial crime detection, value transfer governance, segregation of duties
Human oversight7Escalation, override acceptance, reviewer support
Memory & knowledge18Memory governance, RAG integrity, knowledge management
Sector-specific180Healthcare, finance, agriculture, defence, and other regulated sectors
Other core governance93Reasoning, alignment, output integrity, fairness
Deployment & lifecycle8Release governance, change management
Total508