Thagard's coherence theory for LegionIO: constraint satisfaction across beliefs, goals, and evidence for brain-modeled agentic AI.
Beliefs are not isolated — they support or contradict each other. This extension builds a constraint network of propositions, where positive constraints (explanatory, deductive, analogical, perceptual, conceptual, deliberative) pull propositions toward mutual acceptance, and negative constraints push them apart. The coherence engine runs iterative constraint satisfaction to maximize global coherence, accepting propositions that cohere well with the network and rejecting those that don't.
client = Legion::Extensions::CognitiveCoherence::Client.new
p1 = client.add_coherence_proposition(content: 'The agent should be transparent', domain: :ethics)
p2 = client.add_coherence_proposition(content: 'Transparency builds trust', domain: :ethics)
p3 = client.add_coherence_proposition(content: 'Transparency risks privacy', domain: :ethics)
client.add_coherence_constraint(
prop_a_id: p1[:proposition_id],
prop_b_id: p2[:proposition_id],
constraint_type: :explanatory,
positive: true
)
client.add_coherence_constraint(
prop_a_id: p1[:proposition_id],
prop_b_id: p3[:proposition_id],
constraint_type: :deliberative,
positive: false
)
client.maximize_coherence
client.find_contradictions
client.cognitive_coherence_statsbundle install
bundle exec rspec
bundle exec rubocopMIT