lex-predictive-processing

Multi-domain predictive processing system for the LegionIO cognitive architecture. Manages per-domain generative models with automatic inference mode selection.

What It Does

Maintains a collection of generative models, one per cognitive domain. Each model tracks prediction accuracy, precision, and free energy. The inference mode adapts automatically: perceptual when free energy is low, active when moderate, and hybrid when high free energy coincides with sufficient precision. Each tick, precision decays across all models.

Usage

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

# Add a domain model
client.add_generative_model(domain: :api_latency)

# Generate a prediction
pred = client.predict_from_model(domain: :api_latency, context: { expected: 0.2 })
# => { predicted: true, domain: :api_latency, prediction: { expected_value: 0.2, ... } }

# Observe what actually happened
result = client.observe_outcome(domain: :api_latency, actual: 0.8, predicted: 0.2)
# => { observed: true, domain: :api_latency, prediction_error: 0.6,
#      inference_mode: :active, free_energy: 0.6, state: :surprised }

# Check inference mode
client.inference_mode(domain: :api_latency)
# => { domain: :api_latency, mode: :active }

# Global free energy
client.free_energy
# => { global_free_energy: 0.3 }

# Domain-specific free energy
client.free_energy(domain: :api_latency)

# Domains needing attention
client.models_needing_update
client.active_inference_candidates

# Periodic tick (decay)
client.update_predictive_processing
client.predictive_processing_stats

Inference Modes

Mode	Condition
`:hybrid`	free_energy > 0.7 AND precision >= 0.5
`:active`	free_energy > 0.5
`:perceptual`	otherwise

Development

bundle install
bundle exec rspec
bundle exec rubocop

License

MIT