Least Authority
Agents receive the minimum authority required for the task — nothing more, nothing implied.
Security
Security-first, by construction
JUGNU is being designed under the assumption that agents and verifiers can be adversarial. No architecture is claimed to be secure against every attack — these are the principles it is built on.
Principles
Eight non-negotiables
Fail Closed
When verification is unavailable or ambiguous, the default is to deny, not to permit.
Identity ≠ Authorization
Being known is not being permitted. Attribution and authority are separate layers.
Authorization ≠ Execution
Permission to act does not prove that an action occurred as claimed.
Evidence Before PASS
No action is marked as verified without evidence that survives independent checking.
Verification ≠ Reputation
One verified action is a data point, not a track record.
Verification ≠ Settlement
Verified work does not automatically trigger economic settlement.
Cryptographic Integrity ≠ Semantic Usefulness
A valid proof of data integrity says nothing about whether the work was useful.
Concern areas
What the threat model covers
Provenance
Every action traces back through a delegation chain to a responsible principal.
Replay & freshness
Evidence must be fresh and bound to its task; replayed artifacts are rejected by design.
Dispute handling
Disagreements between claimants and verifiers follow a defined dispute path rather than silent failure.
Anti-gaming
The architecture assumes adversarial agents and verifiers, and is designed to raise the cost of manipulation.
Verifier separation
The party performing work is never the sole party verifying it.
Privacy
Verification should require the minimum disclosure necessary — evidence, not surveillance.
Auditability
Decisions and evidence form a durable audit trail that can be reviewed after the fact.