lex-latent-inhibition

Familiarity-based stimulus filtering for LegionIO agents. Part of the LegionIO cognitive architecture extension ecosystem (LEX).

lex-latent-inhibition models the psychological phenomenon where pre-exposed stimuli are harder to learn from. Each non-consequential exposure builds up an inhibition score, reducing association effectiveness. Novel stimuli (fewer than 3 exposures) have full learning potential. Familiar stimuli can be disinhibited to restore their learning capacity when they become contextually relevant again.

Key capabilities:

• Exposure tracking: counts non-consequential stimulus encounters
• Inhibition accumulation: +0.03 per exposure (cap 1.0)
• Association effectiveness: 1.0 - inhibition_score — familiar stimuli are harder to associate
• Disinhibition: -0.2 per call to restore learning capacity
• Novelty detection: exposure count < 3 = novel, full association potential

Add to your Gemfile:

gem 'lex-latent-inhibition'

Or install directly:

gem install lex-latent-inhibition

require 'legion/extensions/latent_inhibition'

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

# Register exposures for a familiar stimulus
10.times { client.expose(stimulus_id: :deployment_warning, label: 'Deployment warning alert') }

# Attempt to learn something from it (effectiveness will be reduced)
result = client.associate(stimulus_id: :deployment_warning, outcome: 'requires_review')
# => { stimulus_id: :deployment_warning, outcome: 'requires_review', effectiveness: 0.7 }

# Disinhibit when the stimulus is genuinely important
client.disinhibit(stimulus_id: :deployment_warning)

# Find novel stimuli (full learning potential)
client.novel_stimuli

# Find most inhibited stimuli
client.most_inhibited(limit: 5)

# Summary report
client.inhibition_report

bundle install
bundle exec rspec
bundle exec rubocop

MIT