lex-relevance-theory

Relevance Theory-based cognitive input filtering for the LegionIO cognitive architecture. Scores and filters inputs by the ratio of cognitive effect to processing effort.

What It Does

Implements Sperber and Wilson's Relevance Theory: the relevance of an input equals its cognitive effect divided by the effort required to process it. Inputs above a relevance threshold are flagged as worth processing; others are filtered out. Cognitive effect decays over time; processing effort inflates (simulating habituation to repeated inputs). Provides attention budget reporting showing what fraction of inputs merit full processing.

Usage

client = Legion::Extensions::RelevanceTheory::Client.new

# Submit a cognitive input
result = client.submit_relevance_input(
  content:           'Critical security event detected on gateway node',
  input_type:        :observation,
  context:           :security,
  cognitive_effect:  0.9,
  processing_effort: 0.3,
  effect_type:       :new_implication,
  source_id:         'mesh_node_42'
)
input_id = result[:input][:id]

# Assess its relevance
client.assess_input_relevance(input_id: input_id)
# => { success: true, relevance: 3.0, normalized: 1.0, label: :maximally_relevant,
#      worth_processing: true, effect: 0.9, effort: 0.3 }

# Strengthen or weaken
client.strengthen_relevance(input_id: input_id, amount: 0.05)

# Attention budget overview
client.attention_budget_report
# => { total_inputs: 10, worth_processing: 6, filtered_out: 4,
#      processing_ratio: 0.6, avg_relevance: 0.52 }

# Most relevant inputs
client.most_relevant_inputs(limit: 5)
client.worth_processing_report

# Context-specific relevance
client.context_relevance_report(context: :security)

# Periodic decay
client.update_relevance_theory
client.relevance_theory_stats

Relevance Labels

:maximally_relevant (>= 0.8), :highly_relevant, :moderately_relevant, :marginally_relevant, :irrelevant

Development

bundle install
bundle exec rspec
bundle exec rubocop

License

MIT