0/100
MULTI-AGENT CLOUD GOVERNANCE / EXPERIMENT 001
Put cloud changes
on trial.
Seven specialist agents review the same architecture independently. A deterministic policy gate holds the veto. Every vote becomes evidence.
Two versions of the same payment API: risky and corrected. Each runs a fresh review; the verdict can vary.
07Independent reviewers
01Deterministic veto
SHA-256Evidence chain
LIVESingle vs council test
THE COUNCIL
SESECURITYAttack surface
IAIDENTITYLeast privilege
NWNETWORKTrust boundaries
SRRELIABILITYFailure modes
FOFINOPSUnit economics
COCOMPLIANCEControl evidence
PAPLATFORMOperability
CCPOLICY HOLDS
THE VETO
THE VETO
SUBMIT A CHANGE
The court will hear it now.
Use a synthetic or sanitized proposal. Never submit credentials, production secrets, customer information or confidential architecture.
THE METHOD
Persuasion is not proof.
Multi-agent systems often sound more rigorous because they produce more discussion. CloudCourt measures whether the council actually changes a decision and preserves the evidence required to inspect why.
Independent reading
Each specialist reviews the same sanitized change before seeing another agent's conclusion.
Policy before prose
Critical controls are evaluated in code. An eloquent model cannot talk the gate out of a veto.
Measured disagreement
The system records whether the full council changes the single architecture agent's initial result.
Evidence by design
Every case event includes the previous event's SHA-256 hash, creating a verifiable chain.