Articles
Papers
| Paper | Authors | Published | DOI |
|---|---|---|---|
| Semantic Tension Language (STL): A Theoretical Framework for Structured and Interpretable Knowledge Representation | Wuko | Nov 2025 | 10.5281/zenodo.17585432 |
Reports
| Article | Topic | Key Finding |
|---|---|---|
| Context Compression for LLMs | LLM Memory | STL achieves 1.76x better token efficiency than auto-compacted NL (up to 10x vs verbose prose) |
| STL Retrieval Robustness in Long-Context LLMs | Long-Context Retrieval | STL achieves +2.8% higher retrieval accuracy than NL across 1,507 trials; +50% on precise numerical data (N06: NL=0.50, STL=1.00) |
| STL as IR for LLM Code Generation | Vibe Coding | STL achieves 68% feature completion vs NL 45% vs STLC 53%; only format with zero human intervention and first-attempt compilation |
| STG as Skill Library | Agent Skills | Unifying declarative and procedural knowledge in a single graph — recall-to-action via propagation, with Hebbian reinforcement and natural deprecation |
Coming Soon
- STL vs JSON for AI Applications — When structured semantics beats key-value pairs
- Building Persistent AI Memory — Cross-session knowledge accumulation with STL
- The Tension-Path Model — Why directional relations matter for knowledge representation