Developers seeking to audit LLM reasoning paths now have a new tool in their stack: the Trinary Nine-Dimensional 19683 Cognitive System. This open-source project introduces a deterministic symbolic coordinate layer that sits atop large language models, addressing structural weaknesses in continuous vector representations such as non-reproducibility and blurry generalization boundaries. By reverse-engineering the symbolic paradigm of the Yijing (I Ching) and migrating it to a trinary nine-dimensional axiom system, the project offers a locatable, recomputable, and auditable alternative for specific cognitive tasks.

The Axiomatic Skeleton

The system locks in a strict mathematical framework comprising three yao statesβ€”yang (+1), zhong (0), and yin (-1)β€”arranged into nine-dimensional vectors. This structure generates a full space of 19,683 unique gua states, calculated as 3^9. The axioms are immutable, meaning humanistic interpretations cannot alter the underlying mathematics. Five specific operators, including Inversion (F_C) and Changing-yao (F_B), allow for dynamic state transitions, while Euclidean distance calculations enable Top-N neighborhood retrieval within the fixed coordinate space.

Four-Layer Cognitive Engine

The engine operates through four distinct services. First, the Coordinate Service maps natural language situations to specific 9-dimensional vectors. Second, the Five-Domain Semantic Bridge connects these coordinates across domains like healing, time-position judgment, and essence-function-change, claiming a 12/12 sampled mechanical consistency. Third, Neighborhood Generalization uses the Euclidean proximity of gua states to transfer knowledge from known situations to adjacent ones. Finally, the LLM Hybridization layer packages these capabilities as an Agent-callable tool registry, allowing models to explicitly invoke localization and cross-domain bridging rather than relying solely on implicit vector associations.

Open-Source Ecosystem and Implementation

The project is distributed across several repositories, including sanjiu-core for axioms and operators, and sanjiu-ecosystem for the semantic bridge and Agent tools. A formal publication is available on Zenodo with DOI 10.5281/zenodo.23437930. The system ships with a 19,683-entry human gua canon, where interpretations are generated from image-meaning skeletons to ensure traceability. This approach aims to prevent the canon from degenerating into free textual creation, maintaining a strict link between symbolic coordinates and human-readable output.

Key Takeaways

  • The system creates a deterministic coordinate space of 19,683 states using trinary nine-dimensional axioms.
  • It addresses LLM weaknesses in auditability and reproducibility by providing explicit symbolic locations for situations.
  • The tool is packaged as an Agent-callable registry with encode, search, bridge, and decide functions.
  • Open-source under MIT/CC-BY-4.0, with a formal Zenodo paper and GitHub-hosted project homepage.

The Bottom Line

This system doesn't replace LLMs but adds a crucial layer of auditability that pure vector spaces lack. For developers who need reproducible reasoning paths, the shift from implicit association to explicit coordinate computation is a significant engineering win.