Zierax/Animals-Axiom-01-v1

Axiom-01: Phonosemantic Reasoning Engine

Overview

Axiom-01 is a white-box hypothesis-driven reasoning system designed to discover latent structures between phonetic profiles and biological/contextual traits. Unlike black-box neural networks, Axiom-01 operates by identifying statistical laws within a dataset, forming explicit hypotheses, and applying those hypotheses to classify, explain, and generate data with full traceability.

The system is currently implemented as a seed for generalizable reasoning, using animal naming conventions (Phonosemantics) as its primary domain. It demonstrates that structured inference can emerge from statistical law discovery without hardcoded linguistic rules.

Core Principles

• White-Box Architecture: Every conclusion is mapped to a specific hypothesis_id with supporting metrics (Pearson/Spearman correlations).
• Deterministic Reasoning: Identical inputs and datasets yield identical reasoning paths and outputs.
• Truthimatics Logic: Prioritizes logic compression and signal-to-structure mapping over massive data ingestion.
• Explicit Uncertainty: The engine identifies and labels low-confidence results and anomalies rather than hallucinating certainty.

System Architecture

1. Phonetic Signal Space (phonetics.py)

Processes raw strings into quantitative phonetic vectors, including:

• Syllable counts (English and Latin)
• Plosive vs. Soft opening ratios
• Vowel heaviness and sibilance scores

2. Hypothesis Engine (hypothesis_engine.py)

The "brain" of the system. It performs:

• Law Discovery: Scans the dataset for correlations between phonetics and biological traits (e.g., danger score, speed, habitat overlap).
• Confidence Scoring: Assigns weights based on statistical significance.
• Anomaly Registration: Flags data points that violate active laws (e.g., historical outliers or linguistic borrowings).

3. Reasoning Engine (reasoning_engine.py)

Provides three operational modes:

• Classify: Predicts biological profiles based on a name's phonetic structure.
• Explain: Deconstructs an existing name to show which hypotheses it supports or violates.
• Generate: Solves for phonetic structures that align with a target biological profile (e.g., generating a "dangerous-sounding" name).

Installation

Ensure you have Python 3.8+ and the required scientific stack installed:

pip install numpy scipy

Usage

The system is controlled via main.py. Below are the primary commands:

Full Demonstration

Run the complete suite of tests, including classification, explanation, and generation:

python main.py

Classification

Predict the biological profile of a name:

python main.py classify "wolf"

Structured Explanation

Explain the reasoning path for a specific entry in the dataset:

python main.py explain "hippopotamus"

Name Generation

Generate candidates based on biological constraints (e.g., high danger, high encounter frequency):

python main.py generate --danger 0.9 --encounter 0.8 --insider

System Summary

View all currently discovered and active hypotheses:

python main.py summary

Data Model

The system utilizes a master dataset (_dataset.py) containing over 230 animals across four categories (wild, bird, marine, amphibian). Each entry includes:

• Quantitative Metrics: Mass, speed, danger, and encounter frequency scores.
• Contextual Metrics: Namer distance (Insider vs. Outsider origin).
• Qualitative Contrast: Documentation of the discrepancy between survival-based naming (English) and scholarly labels (Latin).

Benchmarking

Axiom-01 includes a comprehensive benchmarking suite (benchmark_report.txt) that evaluates:

• Hypothesis Stability: 100-resample bootstrap robustness.
• Prediction Residuals: RMSE for syllable counts and MAE for back-prediction.
• Topology of Violations: Analysis of why certain names (e.g., "hippo") act as distance anomalies.

Technical Constraints

• Zero Hardcoding: No predefined mappings (e.g., "short = fast") exist. All logic is emergent.
• Metric Floor: A hypothesis is only registered if $|r| \geq 0.20$ and is consistent across $\geq 70%$ of the dataset.
• Z-Score Gating: Anomalies are flagged when an entry exceeds a $1.8$ Z-score for a specific hypothesis.

Identity Statement

Axiom-01 is a mathematical auditor of structure. It does not "feel" the data; it calculates the underlying mind or mechanism behind the signal. It serves as a prototype for interpretable AI where the goal is not just accuracy, but documented understanding.