lex-argument-mapping

Toulmin-model argument mapping for LegionIO — structured claim construction, strength assessment, and soundness evaluation.

Models Stephen Toulmin's argumentation framework: claims supported by grounds, justified by warrants, reinforced by backing, hedged by qualifiers, and challenged by rebuttals. The agent can construct formal arguments for its decisions, assess their logical strength, and identify where rebuttals weaken the case.

An argument has six components:

• Claim: the conclusion being argued for
• Grounds: factual evidence supporting the claim
• Warrant: the logical principle connecting grounds to claim
• Backing: support for the warrant itself
• Qualifier: epistemic hedge (certainly/presumably/probably/possibly/apparently)
• Rebuttal: conditions under which the claim would not hold

strength = (grounds_count * 0.3) + (warrant_present * 0.3) + (backing_count * 0.2)
         - sum(rebuttal_penalties)

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

# Build a Toulmin argument
arg = client.create_argument(
  claim: 'we should migrate to Vault for secrets management',
  domain: :architecture,
  warrant: 'centralized secrets reduce credential sprawl',
  qualifier: :presumably
)

# Add supporting components
client.add_argument_ground(argument_id: arg[:id], ground: 'current rotation is manual and error-prone')
client.add_argument_backing(argument_id: arg[:id], backing: 'CIS benchmark recommends centralized secrets')

# Add a rebuttal
client.add_argument_rebuttal(argument_id: arg[:id], content: 'Vault adds operational complexity', impact: 0.4)

# Assess strength
client.assess_argument_strength(argument_id: arg[:id])
# => { strength: 0.72, label: :strong, rebuttal_count: 1 }

# Find strongest arguments
client.strongest_arguments_report(limit: 3)

Wire into lex-tick's action_selection phase to require formal justification before high-stakes actions. Feed sound_arguments_report into lex-governance proposals for council review.

bundle install
bundle exec rspec
bundle exec rubocop

MIT