HSCI — Hyper-Symbolic Cognitive Invention
An experimental self-verifying cognitive architecture coupling neural search heuristics with Microsoft Z3 SMT logic solvers, structured AST WorkingMemory, and zero third-party commercial LLM dependencies.
[PROJECT STATUS: RESEARCH / EXPERIMENTAL]
HSCI v0.1.0-alpha is a self-contained cognitive architecture backed by 206 passing `pytest` tests. It avoids 3rd-party commercial LLM APIs, utilizing local matrix operations and formal SMT theorem proving.
HSCI v0.1.0-alpha is a self-contained cognitive architecture backed by 206 passing `pytest` tests. It avoids 3rd-party commercial LLM APIs, utilizing local matrix operations and formal SMT theorem proving.
1. Executive Overview
Unlike probabilistic language models that estimate tokens statistically, HSCI executes reasoning cycles through Axiomatic Deliberation. It converts user queries into structured semantic AST frames, spreads activation over concept networks, verifies logical constraints using Microsoft Z3 SMT solvers, and outputs explainable answers.
2. System Architecture
HSCI Component Execution Pipeline
flowchart TD
A[Input Query] --> B[Understanding Engine]
B --> C[Knowledge Manager Facade]
C --> D[Concept Activation Engine CAE]
D --> E[Working Memory AST Frames]
E --> F[Cognitive Reasoning Engine CRE]
F --> G[NativeSymbolicEngine Z3 SMT]
G -->|SAT: Valid| H[Answer Generation Engine AGE]
G -->|UNSAT: Contradiction| I[SQLite SAVEPOINT Rollback]
I --> F
H --> J[Explainable Answer Output]
3. Key Subsystems
- BrainKernel (Core): Manages initialization, stage transitions, and thread-isolated Z3 solver cleanups. (Startup: 1.82ms).
- WorkingMemory (State): Holds semantic frames and attention snapshots backed by SQLite `SAVEPOINT` rollbacks.
- NativeSymbolicEngine (Logic): SMT-backed axiom validator evaluating logic satisfiability via Microsoft Z3.
- Concept Activation Engine (CAE): Spreads activation values over conceptual graphs with decay algorithms.
- Evaluation Framework: Automated `evaluation_runner.py` testing `Basic_Math.json`, `Java_OOP.json`, and `Logic.json`.
4. Technology Stack by Role
| Layer / Role | Technologies Used |
|---|---|
| Core Language & Runtime | Python 3.11+, Asyncio |
| Theorem Proving & Logic | Microsoft Z3 Theorem Prover (SMT Solver) |
| State & Persistence | SQLite (Transactional SAVEPOINTs), JSONL Episode Storage |
| Testing & CI/CD | pytest (206 passed tests), GitHub Actions CI |