Toulmin argumentation engine — symbolic validation of argument structure, inferential integrity, and audience-calibrated delivery.
"Give me a place to stand, and I will move the world." — Archimedes
Rhetoric is a symbolic argumentation engine based on Stephen Toulmin's model. It decomposes natural-language arguments into formal graphs (claim, data, warrant, backing, qualifier, rebuttal), validates inferential steps through four deterministic passes, selects delivery strategy based on audience epistemic state, and optionally generates structural analogies via Mercury.
plan "<intent>" [--no-bridge] [--contradict <evidence>...]Decompose a natural-language argument into a Toulmin graph, validate it through four passes, select a delivery strategy, and optionally generate a structural analogy via Mercury.
python3 rhetoric/scripts/rhetoric.py plan "PostgreSQL 16 improves performance for most workloads"
python3 rhetoric/scripts/rhetoric.py plan "Rust is best for all backends" --contradict "Compile times are slow" "Python has larger ecosystem"
python3 rhetoric/scripts/rhetoric.py plan "Microservices improve reliability" --no-bridge
Requires: INCEPTION_API_KEY environment variable (Mercury API)
Options:
--no-bridge: Skip analogy bridge generation--contradict <evidence>...: Known contradicting evidence for cross-reference validationvalidate <file>Validate a pre-built Toulmin argument graph from a JSON file. No API needed.
python3 rhetoric/scripts/rhetoric.py validate path/to/graph.json
demoRun built-in demonstration cases showcasing the validation engine. No API needed.
python3 rhetoric/scripts/rhetoric.py demo
Demonstrates detection of: valid induction, hasty generalization, formal fallacies (affirming the consequent), cherry-picking (cross-reference integrity), and surface analogies.
The engine runs four deterministic validation passes (no LLM calls):
| Pass | Name | Detects |
|---|---|---|
| 1 | Structural completeness | Missing claim/data/warrant, absent backing/qualifier/rebuttals |
| 2 | Inferential type validation | Formal fallacies, hasty generalization, false cause, surface analogy |
| 3 | Cross-reference integrity | Unaddressed contradictions, unused supporting evidence |
| 4 | Qualifier calibration | Overclaim/underclaim relative to evidence strength |
INCEPTION_API_KEY=... # Mercury API — only needed for 'plan' command
The validate and demo commands require no API keys.
See constitution.md for 6 inviolable rules governing the argumentation engine.