Show HN: ER Intern Simulator Built in 3 Hours via Natural Language Rules (github.com)

An independent researcher released NLCS (Natural Language Constraint System), a repository and whitepaper proposing a method to turn LLMs from probabilistic text generators into deterministic “Semantic Engines” by encoding structured narrative rules as executable constraints. The project includes a math-backed whitepaper (English and Korean), browser-run proof-of-concept simulators — a Combat Simulator v8.4 (real-time combat logic and “vector gravity” fields) and a Growth Simulator v2.0 (economic balance and long-term resource planning) — plus claimed evaluations by GPT-5.1, Claude Opus 4.5 and Gemini 3.0. The author frames Natural Language Programming (NLPg) as using narrative structure itself as source code; the repo is MIT-licensed and includes theoretical proofs of the approach. Technically, NLCS asserts that natural-language rules can create a “vector gravity field” in embedding space that attracts model outputs to specific logical outcomes, and that accumulating rules cause “margin collapse,” reducing hallucination toward zero. If validated, this suggests a pathway for non-code orchestration of model behavior, improved determinism, and tighter safety/control mechanisms for agentic systems. Key implications for the AI/ML community include new approaches to model steering, reproducibility of complex behaviors, and potential trade-offs: dependence on embedding geometry, brittleness to phrasing, and the need for independent empirical validation beyond model-generated endorsements.