Semantic Tension Language

A calculable language for structured knowledge — and the trust layer for delegated AI work. When nobody is watching your agent, its records are all you have. STL makes them contracts.

See a real delegation, audited

🛡

Trust-Layer Protocol Suite

Three open specs for delegated agent work: Intent Contracts (misread intent cannot acquire authority), Board-as-Spec (plans that survive their planner), Eval-Verdict (verification discipline from a year of autonomous-operation scars). Validated by cross-vendor clean-room runs.

👁

Human-Readable

STL syntax is intuitive: [Source] -> [Target] ::mod(confidence=0.9). Anyone can read it. Any LLM can understand it. No parser required for comprehension.

🛡

Traceable & Verifiable

Every statement can embed provenance, confidence scores, and temporal context. Know where knowledge came from and how much to trust it.

⚙

Calculable Tension

Tension is STL's unit of meaning: the directed, weighted pull between two concepts. Statements form tension paths you can compute over — semantic distance, contradiction detection, activation propagation through knowledge graphs.

⚡

Token-Efficient

Compact syntax optimized for LLM context windows. Encode more knowledge in fewer tokens than JSON, XML, or natural language.

See STL in Action

Basic Statement

[Einstein] -> [Theory_Relativity] ::mod(
  rule="logical",
  confidence=0.99,
  domain="physics",
  source="doi:10.1002/andp.19053221004"
)

Causal Chain

[Heavy_Rain] -> [Urban_Flooding] ::mod(
  rule="causal",
  confidence=0.85,
  strength=0.80
)

[Urban_Flooding] -> [Road_Closure] ::mod(
  rule="causal",
  confidence=0.90,
  strength=0.85
)

The Trust Layer for Delegated AI Work

Three self-contained specifications an agent can execute from the text alone — plus the exhibit that shows one doing exactly that, traps and all.

Read the Trust-Layer Suite

This website is AI-native. All content is authored in .stl files and served directly to AI agents at /.well-known/site.stl