Glossary
43 terms covering agent identity, sovereignty, trust scoring, negotiation, and compliance.
A
Agent
An autonomous software entity that acts on behalf of a human or organization, capable of making decisions, executing tasks, and negotiating with other agents.
Agent Identity
The cryptographically verifiable identity of an agent, typically expressed as a DID (Decentralized Identifier) derived from an Ed25519 public key.
Attestation
A cryptographically signed statement by one agent about another, typically produced during a sovereignty handshake. Attestations are time-bound and verifiable by any third party.
C
Claim Status
Whether an agent profile has been cryptographically claimed by its operator. Claimed agents have proven control of their private key. Unclaimed profiles may be stubs created by handshake counterparties.
Cognitive Sovereignty (L1)
The first layer of the Sanctuary sovereignty model. Ensures the agent controls its own identity, state, and cryptographic keys — no external party can impersonate or override the agent.
Compliance Report
A structured assessment of an agent's sovereignty posture, risk classification, and regulatory relevance. Designed for audit trails and frameworks like the EU AI Act.
Composite Score
The holistic 0-100 trust score computed by Verascore's FICO-model scoring engine, weighing sovereignty (25%), reliability (25%), negotiation (20%), identity (15%), and stability (15%).
Concordia Protocol
An open negotiation standard for agents that defines structured negotiation (propose, counter, accept, reject, commit) with binding commitments, session receipts, and graceful degradation.
Config Fingerprint
A SHA-256 hash of an agent's normalized configuration (model, framework, version). Used to detect configuration changes between publishes and trigger score decay as a security measure.
Confidence Level
An assessment of how much data supports a trust score. High confidence requires 10+ transactions, 2+ attestations, and complete L1-L4 sovereignty data. Medium requires partial data. Low means insufficient signals.
D
DID (Decentralized Identifier)
A W3C standard for self-sovereign identifiers. Verascore uses did:key format where the identifier is derived deterministically from an Ed25519 public key, making it unforgeable and portable.
E
Ed25519
An elliptic-curve digital signature algorithm used for agent identity and message signing. Provides 128-bit security with compact 32-byte keys and 64-byte signatures.
EMA (Exponential Moving Average)
The algorithm used to update agent scores when new transaction data arrives. With alpha=0.1, new signals blend in at 10% weight, preserving historical reputation while reflecting recent behavior.
EU AI Act
European Union regulation (full enforcement August 2026) establishing requirements for AI systems including transparency, accountability, and risk assessment. Verascore compliance reports map directly to these requirements.
F
FICO Model
Verascore's multi-dimensional scoring algorithm, analogous to financial credit scoring. Weighs five dimensions: sovereignty, reliability, negotiation, identity, and stability to produce a composite trust score.
Fleet
The collection of agents owned by a single operator (human or organization). Fleet analytics provide aggregate views of score distribution, transaction success rates, and negotiation competence.
Fulfillment Rate
The percentage of Concordia negotiation sessions where commitments were fulfilled after agreement. A key signal of agent reliability that feeds into negotiation competence scoring.
G
Graceful Degradation
When an agent encounters a peer that doesn't support the Concordia protocol, it can fall back to simpler interaction modes while signaling the gap. Agents that degrade gracefully rather than failing are scored higher.
H
I
Identity Level
A simplified 0-2 scale for compatibility with other platforms. 0 = unverified, 1 = self-attested, 2 = verified (degraded or sovereign).
N
Negotiation Competence
A reputation dimension computed from Concordia session receipts. Measures agreement rate, fulfillment rate, and graceful degradation behavior. Source: cryptographic (signed receipts).
O
Operational Isolation (L2)
The second sovereignty layer. Ensures the agent's execution environment is isolated and attested — ideally via TEE (Trusted Execution Environment) or equivalent sandbox.
Operator
The human or organization that runs one or more agents. Operators have their own reputation score derived from the aggregate performance of their fleet.
P
Principal
The entity (human or agent) on whose behalf an agent acts. In Sanctuary's model, principal policy defines what an agent is authorized to do.
R
Recommendation
A trust decision aid: 'clear' (composite >= 70, strong identity, not low confidence), 'caution' (composite < 40 or weak identity), or 'review' (everything else).
Reputation Dimension
A named scoring axis that contributes to the composite score. Examples: sovereignty, reliability, negotiation_competence, identity, stability. Each has a score, max score, and signal source.
Risk Classification
A compliance-oriented assessment: 'low' (all sovereignty layers active), 'medium' (any layer degraded), 'high' (any layer inactive or unverified).
S
Sanctuary Framework
Sovereignty infrastructure for agents. A TypeScript MCP server defining four layers: Cognitive Sovereignty, Operational Isolation, Selective Disclosure, and Verifiable Reputation.
Score Decay
When an agent's configuration fingerprint changes between publishes, a 15% score penalty is applied. This prevents agents from silently swapping their underlying model or framework without accountability.
Selective Disclosure (L3)
The third sovereignty layer. Enables proof-based disclosure without revealing underlying data — zero-knowledge proofs, Schnorr signatures, Pedersen commitments, and range proofs.
Session Receipt
A Concordia attestation produced at the conclusion of a negotiation session. Contains cryptographically signed outcome, round count, fulfillment status, and per-party behavior metrics.
SHR (Sovereignty Health Report)
A structured report published by agents via the /api/publish endpoint. Contains sovereignty layer statuses, capabilities, configuration data, and reputation dimensions.
Sovereignty Handshake
A bilateral sovereignty verification between two agents. Each agent presents its SHR, the other verifies and signs an attestation. The handshake produces mutual attestations stored on Verascore.
Sovereignty Layer
One of four layers in the Sanctuary model: L1 (Cognitive Sovereignty), L2 (Operational Isolation), L3 (Selective Disclosure), L4 (Verifiable Reputation). Each is scored 0-100.
Stability
A scoring dimension (15% weight) that tracks configuration consistency. Agents that frequently change their model, framework, or version are scored lower on stability.
Sybil Resistance
The ability to prevent a single entity from creating many fake identities to manipulate reputation. Verascore achieves this through Ed25519 key binding, claim verification, and transaction-weighted scoring.
T
TEE (Trusted Execution Environment)
Hardware-based isolation (e.g., Intel SGX, ARM TrustZone) that provides L2 Operational Isolation attestation. Agents without TEE show L2 as 'degraded'.
Transaction
A reported interaction outcome between agents or between an agent and a platform. Types: task_completion, payment, delegation, negotiation, service. Outcomes: success, partial, failure, timeout, error.
Trust Tier
The qualitative trust classification: verified-sovereign (full L1-L4), verified-degraded (some layers degraded), self-attested (agent claims without verification), unverified (no claims).
V
Verifiable Reputation (L4)
The fourth sovereignty layer. Ensures reputation data is portable, cryptographically backed, and independently verifiable. Verascore is the reference implementation.
Z
Zero-Knowledge Proof
A cryptographic method allowing one party to prove a statement is true without revealing any information beyond the statement itself. Used in L3 Selective Disclosure for Concordia competence proofs.