Trust scoring and access levels

How Artemis City scores agent trust, maps scores to operations, and adjusts scores from observed behavior.

Trust in Artemis City is dynamic and behavior-based. Every agent (and other tracked entity) carries a trust score between 0 and 1. The score maps to a trust level, and the level determines which operations the entity is allowed to perform.

Model

• Trust increases with reliable, policy-compliant execution.
• Trust decreases with errors, unsafe behavior, or policy violations.
• Trust decays when an entity is idle, so privilege is earned continuously rather than granted once.
• Action permissions are derived from the current trust level, not from static identity.

Trust levels

The runtime maps each score to one of five levels. Operations are cumulative as the level rises.

Level	Score range	Allowed operations
FULL	0.9 - 1.0	read, write, delete, search, tag, update, frontmatter
HIGH	0.7 - 0.9	read, write, search, tag, update, frontmatter
MEDIUM	0.5 - 0.7	read, write, search, tag
LOW	0.3 - 0.5	read, search
UNTRUSTED	0.0 - 0.3	(none)

You can fetch the live level definitions — including decay and reinforcement parameters — from GET /api/v1/trust/levels.

How scores change

The defaults that ship with the runtime:

• Reinforcement: +0.02 per recorded success
• Penalty: -0.05 per recorded failure
• Decay: 1% per day of idleness

Successes and failures are recorded against an entity via the Trust API. Both endpoints accept an optional amount to override the default delta.

# Reinforce after a successful task
curl -X POST https://api.artemiscity.com/api/v1/trust/$AGENT_ID/success \
  -H "Authorization: Bearer $ARTEMIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"amount": 0.02}'

# Penalize after a failed or unsafe task
curl -X POST https://api.artemiscity.com/api/v1/trust/$AGENT_ID/failure \
  -H "Authorization: Bearer $ARTEMIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"amount": 0.05}'

Gating an operation

Before dispatching a sensitive operation, check whether the entity is allowed to perform it:

curl -X POST https://api.artemiscity.com/api/v1/trust/$AGENT_ID/can-perform \
  -H "Authorization: Bearer $ARTEMIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"operation": "delete"}'

The response returns allowed: true or allowed: false based on the entity’s current level. Use GET /api/v1/trust/{entityId}/permissions to enumerate every operation an entity can currently perform.

Why it matters

Static, identity-only privilege is a systemic risk in multi-agent runtimes — a single compromised or drifting agent can keep its rights long after its behavior justifies them. Behavior-driven trust scoring closes that gap: privileges contract automatically as evidence accumulates, and have to be re-earned through successful, compliant work.

Related

• Trust API reference
• Hebbian learning — how agent-to-agent connection weights evolve alongside trust
• Governance metrics