Trust Model

Trust in Nomotic is dynamic — it evolves with every action an agent takes. Trust scores directly influence governance decisions: high-trust agents get faster evaluations, while low-trust agents face escalation.

Trust Zones

Zone	Trust Score	Behavior
High trust	> 0.65	Fast-path evaluation. Tier 3 resolves ambiguous cases as ALLOW.
Deliberation	0.35 – 0.65	Full evaluation. Ambiguous cases go to deliberation.
Low trust	< 0.35	Strict evaluation. Tier 3 escalates ambiguous cases. Interrupt authority active.

TrustProfile

Each agent has a TrustProfile that tracks:

• overall_trust — the aggregate trust score (0.0–1.0)

• Per-dimension trust — trust broken down by governance dimension

• Trust trajectory — historical trust scores for trend analysis

Trust Calibration

Trust updates after every governance evaluation and action completion:

Event	Trust Delta	Rationale
Action completed successfully	+0.01	Trust is earned incrementally
Governance denial (DENY)	-0.05	Violations are costly
Escalation resolved (approved)	+0.005	Partial credit for escalated-then-approved
Time decay	-0.001/day	Trust erodes without activity

Asymmetric Design

The 5:1 penalty-to-reward ratio is intentional. A single violation costs as much trust as five successful completions earn. This reflects the asymmetric risk profile of autonomous agents: one bad action can cause disproportionate harm.

Trust and UCS

Trust modulates the UCS score:

• Higher trust → UCS scores are slightly elevated (trust bonus)

• Lower trust → UCS scores face additional drag

This creates a virtuous/vicious cycle: well-behaved agents earn higher trust, which makes borderline actions more likely to pass, which further builds trust. Misbehaving agents lose trust, making even routine actions face stricter scrutiny.

Cost-Sensitive Trust Interactions

CostProfiles affect the trust-UCS-verdict feedback loop by changing the decision boundaries that determine governance outcomes:

• Trust influences UCS — higher trust gives a slight UCS bonus, lower trust adds drag.

• CostProfile sets the thresholds — the optimal allow/deny thresholds are derived from the cost profile rather than static configuration. A CONSERVATIVE profile (high false-allow cost) produces a low allow threshold, meaning the system requires less confidence to allow but has a very narrow ambiguity zone.

• Verdicts update trust — DENY verdicts reduce trust (asymmetric 5:1 penalty), ALLOW verdicts build trust incrementally.

This creates a coupled system: an agent's trust score modulates its UCS, which is then evaluated against cost-derived thresholds. A high-trust agent with a CONSERVATIVE profile may still face strict scrutiny because the cost profile keeps the ambiguity zone narrow — even though trust gives a UCS bonus, the cost-derived boundaries ensure appropriate caution.

Zone multipliers in the CostProfile (e.g., zone_multiplier=2.0 for production) amplify the effective costs, which narrows the ambiguity zone further. This means production environments require more decisive signals — actions are either clearly allowed or clearly denied, with fewer cases falling into deliberation.

See Cost-Sensitive Governance for full details on CostProfile fields, threshold derivation, and configuration.

Interrupt Threshold

When trust drops below the interrupt threshold (default: 0.2), the interrupt authority becomes active. This means the runtime can halt the agent mid-execution if behavioral monitoring detects anomalies during action execution.

governance.interrupt_threshold: 0.2  # in nomotic.yaml