lex-trust

Emergent domain-specific trust modeling for brain-modeled agentic AI. Tracks multidimensional trust scores per agent-domain pair with reinforcement, decay, and delegation thresholds.

Overview

lex-trust implements the agent's model of trust in other agents. Trust is not a single number — it is measured across four dimensions (reliability, competence, integrity, benevolence), is domain-specific, and emerges from observed interactions rather than being assigned. Negative interactions have a larger impact than positive ones (asymmetric update).

Trust Dimensions

Dimension	Description
`reliability`	Does the agent do what it says?
`competence`	Does the agent achieve good outcomes?
`integrity`	Does the agent act with honesty?
`benevolence`	Does the agent act in good faith?

The composite score is the mean of all four dimensions.

Trust Thresholds

Threshold	Value	Meaning
Starting trust	0.3	Neutral starting point for new agents
Consider threshold	0.3	Minimum to include agent input
Delegate threshold	0.7	Minimum to delegate actions
Decay rate	0.005/cycle	Trust decays without interaction
Reinforcement	+0.05/positive	Per positive interaction
Penalty	-0.15/negative	Per negative interaction (asymmetric)

Installation

Add to your Gemfile:

gem 'lex-trust'

Usage

Recording Interactions

require 'legion/extensions/trust'

# Record a positive interaction (increments all 4 dimensions by 0.05)
result = Legion::Extensions::Trust::Runners::Trust.record_trust_interaction(
  agent_id: "agent-42",
  positive: true,
  domain: :task_execution
)
# => { agent_id: "agent-42", domain: :task_execution, positive: true,
#      composite: 0.35, interactions: 1 }

# Record a negative interaction (decrements all 4 dimensions by 0.15)
Legion::Extensions::Trust::Runners::Trust.record_trust_interaction(
  agent_id: "agent-42",
  positive: false,
  domain: :task_execution
)

Querying Trust

# Get trust entry for specific agent/domain
Legion::Extensions::Trust::Runners::Trust.get_trust(agent_id: "agent-42", domain: :task_execution)

# List all agents above consider threshold
Legion::Extensions::Trust::Runners::Trust.trusted_agents(domain: :task_execution)

# List agents above delegate threshold (0.7)
Legion::Extensions::Trust::Runners::Trust.delegatable_agents(domain: :task_execution)

Reinforcing Specific Dimensions

# Reinforce a single trust dimension
Legion::Extensions::Trust::Runners::Trust.reinforce_trust_dimension(
  agent_id: "agent-42",
  dimension: :reliability,
  domain: :task_execution,
  amount: 0.05
)

Decay

# Apply decay to all trust entries (called periodically)
Legion::Extensions::Trust::Runners::Trust.decay_trust
# => { decayed: 3 }  (number of entries updated)

Development

bundle install
bundle exec rspec
bundle exec rubocop

License

MIT