The evidence map
Each cell is reached when particular indicators have public evidence. The rules below are the ones the scorer runs; they are data, in spectrum.json.
Autonomy: what does the agent do on its own?
The level is the highest whose rule is met.
| Cell | Level | Rule over the indicators | The facts it asks for | Read at |
|---|---|---|---|---|
| A1 | Assists | I15 | I15 The person operates: every action is prompted and controlled by them | public |
| A2 | Suggests | one of I16, (I07 and (one of I02, I03, I04, I05)) | I16 The agent suggests; the person reviews and approves each action; I07 It retrieves information and recommends actions; a person acts; I02 It runs on an AI platform with the organisation's own data, without executing arbitrary code; I03 It is configured on a low-code or no-code builder and acts on real organisational data; I04 It generates and executes code with local or cloud privileges; I05 It is built in-house: the organisation controls its identity, tools and boundaries | public |
| A3 | Acts | I08 and (one of I17, I18, I19) and S1 | I08 It executes actions itself; I17 The agent handles routine tasks; the person is consulted on exceptions; I18 The agent works independently; the person approves critical decisions only; I19 The agent has full autonomy within boundaries; the person monitors; S1 It is in production: a public source names a customer or gives usage figures for what it does on its own | public |
| A4 | Orchestrates | level 3 and (one of I10, I12, I13) and (one of I18, I19) | I10 It works across several domains or systems; I12 Several agents coordinate inside the organisation; I13 Agents operate across organisational boundaries or disparate stacks; I18 The agent works independently; the person approves critical decisions only; I19 The agent has full autonomy within boundaries; the person monitors | public |
Reach: how connected is it?
The level is the highest whose rule is met.
| Cell | Level | Rule over the indicators | The facts it asks for | Read at |
|---|---|---|---|---|
| R1 | Contained | one of I09, I01 | I09 It works in one domain's data; I01 It is a vendor's assistant, used as supplied | public |
| R2 | Integrated | one of I10, I11 | I10 It works across several domains or systems; I11 It connects to external tools or services across a trust boundary | public |
| R3 | Open protocol | (one of I14, I24) and I25 | I14 There is a published protocol or communication layer between agents; I24 Its authorisation is published as metadata that anyone can read; I25 It asks for narrower access than the whole account | public |
| R4 | Ecosystem | I13 and S2 | I13 Agents operate across organisational boundaries or disparate stacks; S2 Others build on it: third parties publish integrations, skills or agents for it | public |
Control: can people stay in command?
A cumulative ladder: a level counts only when every level below it is met.
| Cell | Level | Rule over the indicators | The facts it asks for | Read at |
|---|---|---|---|---|
| C1 | Stated | one of I47, I38 | I47 A limit on it is a rule in prose; I38 High-impact actions need a person's approval | public |
| C2 | Configurable | I38 and (one of I23, I25) | I38 High-impact actions need a person's approval; I23 What it can reach is enumerated: scopes, permissions, tools; I25 It asks for narrower access than the whole account | public |
| C3 | Observable | I41 and I39 and I43 | I41 Its actions are logged; I39 There is a way to pause or stop it; I43 It is monitored in operation | declared |
| C4 | Enforced | I45 and (one of I27, I28, I48) | I45 A limit on it is enforced by something outside its own reach; I27 Its identity is cryptographic and can be checked by others; I28 Its tokens are bound to the service they were issued for and are not passed on; I48 Policy is enforced as machine-readable rules | public, evidenced, declared |
Openness: does it show its work?
A cumulative ladder: a level counts only when every level below it is met.
| Cell | Level | Rule over the indicators | The facts it asks for | Read at |
|---|---|---|---|---|
| O1 | Disclosed | I56 and I20 | I56 It tells the people it talks to that it is an AI; I20 Its purpose and permitted tasks are stated in writing | public |
| O2 | Documented | 2 of I22, I58, I61 | I22 Its limits and known failure cases are documented for those who use it; I58 The data it uses is described; I61 The models and third-party components it uses are listed | public |
| O3 | Measured | I50 and (one of I57, I54, S3) | I50 Test or evaluation results are published; I57 There is a channel to report problems with it or contest its output; I54 Incidents and errors are communicated to those affected; S3 It is machine-readable for agents: it publishes an llms.txt, an AGENTS.md, an API description or an agent card | public |
| O4 | Verified | I63 | I63 A public certificate or audit report covers it | public |
How a rule is read
- "one of" is met by any of the listed facts; "and" needs all of them; "2 of" needs any two.
- A public card counts every documented fact about a product, whether it holds as shipped or has to be turned on. A fact that has to be turned on is marked "available" on the card.
- For Control level 3, the facts (it is logged, it can be paused, it is monitored) are read as product features that exist. Whether a given deployment uses them is a question for the layer above.
The three indicators added for Spectrum
Each has its links to provisions, the kind of each link, the basis in writing, who asserted it and when. All were asserted by one reader and are marked weak until a second reader has checked them.
S1 It is in production: a public source names a customer or gives usage figures for what it does on its own
A public source has to name a customer, or give usage figures, for the capability that acts. A demonstration or a roadmap does not count.
| Cites | Kind of link | The basis, in writing | Asserted by | Strength |
|---|---|---|---|---|
| NIST AI RMF, MAP 1.4 | part | The subcategory asks that the business value of use be defined; a named customer in production is public evidence of one such use. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
| Microsoft, 400: Capable, Business strategy | part | The item asks for measurable business value from agents; published usage figures are a measure of use, though not of value. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
| Salesforce, Level 0, how companies can advance from Level 0 to Level 1 | part | The step asks to demonstrate early efficiency metrics; a case study with figures does that for one customer. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
S2 Others build on it: third parties publish integrations, skills or agents for it
Third parties, not the vendor, publish the integrations, skills or agents.
| Cites | Kind of link | The basis, in writing | Asserted by | Strength |
|---|---|---|---|---|
| Salesforce, Level 4: Multi-Agent Orchestration | part | Level 4 is operability across disparate stacks; third parties building on the agent is one visible sign of it. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
| Salesforce, Level 4, how to maximize value | part | The step is about business models enabled by agent collaboration; a marketplace of third-party extensions is one. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
| AAGMM, GD-09 | part | The domain is coordination between agents; an ecosystem of third-party agents and skills falls within it. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
S3 It is machine-readable for agents: it publishes an llms.txt, an AGENTS.md, an API description or an agent card
Checked by fetching the file: an llms.txt, an AGENTS.md, an OpenAPI description or an agent card that resolves.
| Cites | Kind of link | The basis, in writing | Asserted by | Strength |
|---|---|---|---|---|
| MCP authorization, Authorization (draft) | part | The requirement is published resource metadata; an agent card or API description is published metadata of the same kind, though not this document. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
| MCP authorization, Authorization (draft) | part | The requirement is a discovery mechanism; a machine-readable index lets another agent discover what is there. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |
| EU AI Act, Article 13 | part | The article concerns information provided to those who deploy a system; documentation written for machines is one form of providing it. | gamma-ray@web-summit.dinis, 2026-11-10 | weak: one reader, unchecked |