A draft standard · Graph-Analyzed Multi-framework Maturity Assessment with Risk-Adjusted Yield
The more an agent is governed, the further it can be trusted to go.
GAMMA-Ray rates one AI agent on two things: how much it does on its own, and how well that is controlled. It asks 21 questions. Each answer needs evidence, and each answer counts towards several published standards at once.
Two levels, one result
Freedom is what the agent does on its own, from 1 (it answers) to 5 (agents working with agents). Governance is how it is controlled, from 0 (nothing stated) to 4 (proven). An agent is held at the highest freedom level that its governance supports. A small agent that is well governed is mature. A powerful one that nobody can show to be controlled is not, yet.
This is RiskMandate's lens, measured: freedom is the agent's reach, governance is its mandate and what stands in the way, and the distance between the level reached and the level held is the gap.
One short list, many standards
Maturity standards for agents ask many of the same things in different words. We cut 9 instruments into 286 provisions, found the observable fact behind each, and grouped those facts into questions. 215 of the 286 provisions (75%) are reached by 21 questions. The rest are listed as outliers: what only one instrument asks.
- The standard: the levels, the tiers and the rules.
- The questions: all 21, with every citation.
- The mashup: the graph, and what would not compress.
- Coverage: how much of each instrument can be read from outside.
- Scans: our own agents first.
Where this is
A draft, v0.0.1. The mapping between instruments is a first pass by one reader and has not been checked by a second. So far it rates nobody but ourselves. What exists and what does not.