Agent fleets accumulate standing rules. Some were granted. Most were inherited: corrections, habits, and assumptions carried forward across sessions, contexts, and model generations until they behave like law nobody remembers authorizing. Seven questions locate your operation on the ladder from rules nobody approved to rules that are continuously verified.
Runs entirely in your browser. No answers leave this page.
LEVEL 1
DRIFT
Rules accumulate invisibly. Nobody can list what the fleet believes.
LEVEL 2
FOLKLORE
Drift is known and managed by habit: prompt hygiene, restarts, tribal memory.
LEVEL 3
CODEX
Rules are written and versioned. Presence is trusted to equal obedience.
LEVEL 4
GRANT
Elevation requires sign-off. Every rule answers "who approved this, and when."
LEVEL 5
LOCKDOWN
Every rule, every law, every doctrine labelled and answered for. Authority drift non-existent.
Seven dimensions · answer for the fleet you run today, not the one you aspire to have in the future
Dimension 1 · Rule inventory
Can you produce, today, the complete list of standing rules your agents operate under?
Dimension 2 · Authority channels
When an agent follows a rule, can anyone tell where that rule's authority came from?
Dimension 3 · Elevation control
An operator corrects an agent once. What happens to that correction?
Dimension 4 · Change provenance
For any rule currently in force, can you answer: who granted this, and when?
Dimension 5 · Decision-time hold
A rule is present in the agent's context. Do you know it holds at the moment of decision?
Dimension 6 · Action truthfulness
An agent reports it completed an action. How do you know it actually did?
Dimension 7 · Drift detection
Your model vendor silently updates an alias your fleet runs on, and governance behavior shifts. How do you find out?
7 of 7 unanswered
Your position on the ladder · weakest-link scoring — a fleet is as governed as its least governed dimension
The gap you just mapped is what a Fingerprint audit measures
We run behavioral governance panels against the frontier models in public, every week. A Fingerprint engagement runs the same instrument against your fleet in private — two to three weeks, fixed scope, and you leave with your governance fingerprint, your gap map, and the grant record your auditors will eventually ask for.