Tyche-core is a logic-first reasoning framework grounded in semiotics, affect theory, and model-theoretic validation. Developed by philosopher and AI contributor Sam Emswiler, Tyche-core replaces probabilistic outputs with verifiable interpretive chains. The system integrates sign-based logic, formal diagrammatics, and modality-anchored heuristics to enable structured thought in environments of ambiguity, uncertainty, contradiction, or even humor. We design and validate systems for reasoning, understanding, and judgments; looking for partners interested in continuing this development.

It has four main parts:

1. A formal language (E2) for verifiable interpretive proofs

2. A semiotic game and game engine for testing stability of interpretations of signs

3. Human-aligned benchmarks for logic, contradiction, and value assessment

4. Visualization tools for modal logic, deontic structure, and decision making

⚙️ Applied Expertise

Our founder is a contributing logic expert for LLM development and Tyche-core is an offshoot of current dissertation work on diagrammatic logic at the University of Miami influenced by Dr. Risto Hilpinen on logic and Dr. Michael Slote on virtue and affect. LLM background includes:
Authoring and validating graduate-level logical problems to enhance LLM fluency, inference depth, and model self-correction. Contributions include: Building benchmark datasets for LLM fine-tuning and evaluation; Detecting deductive failure modes in AI-generated reasoning ; Engineering prompts and metrics for valid solution generation; Shaping the criteria for rigor, verifiability, and interpretability in machine reasoning.

📈 Potential Use & Integration

AI reasoning evaluation tools

Interpretability frameworks for safety & ethics

Design for logic, ethics, and cognition

Custom dataset development for language models

Developing value-affect “driven” N.P.C.’s or bots for more interactive game or social structure

🚧 Why Tyche-Core?

No one else is trying to build a system in this way with these tools. LLMs today operate through patterns. We are building for meaning. Tyche-core does not just grade reasoning. It generates the conditions under which reasoning can be meaningfully interrogated—by humans or machines. When other systems output answers, Tyche-core builds interpretive structure

Where black-box models produce probabilities, we produce explanations and accountability.